Reset repo for phased rebuild and scaffold Phase 1 Parakeet API

.gitignore

@@ -1,9 +1,11 @@
-*.mp3
-
 __pycache__/
 *.pyc
-benchmark_outputs/
-results.json
+.pytest_cache/
+.venv/
 
 .DS_Store
+
+*.mp3
+*.wav
 *.pdf
+results.json

README.md

@@ -1,5 +1,5 @@
 ---
-title: Transcribe Diarize
+title: Transcribe Diarize Rebuild
 emoji: ⚡
 colorFrom: indigo
 colorTo: indigo
@@ -9,165 +9,71 @@ python_version: "3.12"
 app_file: app.py
 pinned: false
 license: mit
-short_description: End-to-end Parakeet + Pyannote transcript pipeline
+short_description: Phase 1 rebuild scaffold - Parakeet-only transcription API
 ---
 
-This Space is optimized for API usage on ZeroGPU.
+# Phase 1 rebuild: Parakeet-only
 
-## Pipeline
-Single production pipeline:
-1. Parakeet transcription (word timestamps) on ZeroGPU
-2. Pyannote Community-1 diarization on ZeroGPU
-3. Merge transcript + diarization (CPU)
+This repository has been reset for a phased rebuild using `plan/rebuild_plan.md` as the source of truth.
 
-Model setup is global/outside `@spaces.GPU` so setup time is not billed to ZeroGPU windows.
+## Current scope
+Only Phase 1 is active:
+- Parakeet-only transcription API
+- CPU-side preload/probing/postprocessing
+- narrow `@spaces.GPU(duration=120)` window around inference only
 
-## API routes
-- `/transcribe`
-- `/diarize`
-- `/run_complete_pipeline`
-- `/get_debug_output`
-- `/get_run_complete_pipeline_schema`
+Not implemented yet:
+- pyannote diarization
+- merge logic
+- combined pipeline
 
-## `/transcribe` inputs
-- `audio_file` (file path from Gradio client upload)
+## Active API
+- `/transcribe_parakeet`
 
-Returns: Parakeet-only transcription JSON, exactly the payload used inside the full pipeline:
+## Request inputs
+- `audio_file`
+- `model_options_json` (optional JSON object)
 
+## Response shape
 ```json
 {
+  "phase": 1,
+  "request_id": "uuid",
   "model": "NVIDIA Parakeet v3",
-  "task": "transcribe",
-  "audio_file": "/tmp/gradio/example.wav",
-  "postprocess_prompt": null,
-  "model_options": {},
-  "zerogpu_timing": {
-    "gpu_window_seconds": 3.12,
-    "inference_seconds": 2.87
-  },
-  "raw_output": {
-    "output": {},
-    "timestamp_hint": "word timestamps available in output.timestamp['word'] when provided by NeMo",
-    "language_hint": "auto",
-    "long_audio_settings": {}
-  },
-  "timestamp_granularity": "word"
-}
-```
-
-## `/diarize` inputs
-- `audio_file` (file path from Gradio client upload)
-- `huggingface_token`
-
-Returns: Pyannote-only diarization JSON, exactly the payload used inside the full pipeline:
-
-```json
-{
-  "model": "pyannote/speaker-diarization-community-1",
+  "model_id": "nvidia/parakeet-tdt-0.6b-v3",
   "audio_file": "/tmp/gradio/example.wav",
   "duration_seconds": 300.0,
-  "model_options": {
-    "hf_token": "***redacted***",
-    "long_audio_chunk_threshold_s": 7200,
-    "chunk_duration_s": 7200,
-    "chunk_overlap_s": 0
+  "execution_plan": {
+    "mode": "single_pass",
+    "chunking": {
+      "implemented": false,
+      "reason": "Phase 1 scaffold keeps chunk planning on CPU but does not split audio yet."
+    }
   },
-  "long_audio": {
-    "chunk_count": 1,
-    "chunk_plan": [{"index": 0, "start": 0.0, "end": 300.0, "duration": 300.0}],
-    "chunked": false
+  "preload": {
+    "cache_hit": true,
+    "load_seconds": 0.0
   },
   "zerogpu_timing": {
-    "gpu_window_seconds": 8.45,
-    "inference_seconds": 8.11
+    "gpu_window_seconds": 3.12,
+    "inference_seconds": 2.87
+  },
+  "wall_clock_timing": {
+    "probe_seconds": 0.03,
+    "preload_seconds": 0.0,
+    "postprocess_seconds": 0.01,
+    "total_seconds": 3.22
   },
   "raw_output": {
-    "chunk_results": [],
-    "stitched": {
-      "speaker_diarization": [],
-      "exclusive_speaker_diarization": []
-    }
-  }
-}
-```
-
-## `/run_complete_pipeline` inputs
-- `audio_file` (file path from Gradio client upload)
-- `huggingface_token`
-
-Returns: merged transcript JSON only, with this structure:
-
-```json
-{
-  "summary": {
-    "diarization_key_used": "exclusive_speaker_diarization",
-    "parakeet_word_count": 1234,
-    "pyannote_segment_count": 42,
-    "turn_count": 39,
-    "assigned_word_count": 1219,
-    "unassigned_word_count": 15
+    "result": {},
+    "runtime": {}
   },
-  "turns": [
-    {
-      "speaker": "SPEAKER_00",
-      "start": 0.0,
-      "end": 12.34,
-      "text": "Good morning and welcome to the earnings call."
-    }
-  ],
-  "transcript_text": "[0.00 - 12.34] SPEAKER_00: Good morning and welcome to the earnings call."
+  "normalized_output": {
+    "text": "...",
+    "word_timestamps": []
+  }
 }
 ```
 
-## `/get_debug_output` inputs
-- `run_id` (optional)
-
-Returns: raw/debug payload for the latest run (or specific run if provided), including:
-- Parakeet raw output + ZeroGPU timing
-- Pyannote raw output + ZeroGPU timing
-- merged transcript payload
-- aggregated timing
-
-## `/get_run_complete_pipeline_schema` inputs
-- none
-
-Returns a machine-readable contract for `/run_complete_pipeline`:
-- `input_schema`
-- `output_schema`
-- `output_example`
-
-## IPython example
-```python
-from gradio_client import Client, handle_file
-
-SPACE = "username/space-name"
-AUDIO_FILE = "Q3-FY26_5min.mp3"
-
-client = Client(SPACE)
-
-# Run Parakeet only
-parakeet_response = client.predict(
-    audio_file=handle_file(AUDIO_FILE),
-    api_name="/transcribe",
-)
-
-# Run Pyannote only
-pyannote_response = client.predict(
-    audio_file=handle_file(AUDIO_FILE),
-    huggingface_token="hf_xxx",
-    api_name="/diarize",
-)
-
-# Run end-to-end pipeline (returns merged transcript)
-merged_transcript = client.predict(
-    audio_file=handle_file(AUDIO_FILE),
-    huggingface_token="hf_xxx",
-    api_name="/run_complete_pipeline",
-)
-
-# Fetch latest debug payload
-debug_payload = client.predict(
-    run_id=None,  # or a specific run_id returned by prior debug call
-    api_name="/get_debug_output",
-)
-```
+## Legacy code
+The previous implementation was archived under `archive/legacy_pre_rebuild_*/` so the rebuild context stays clean.

app.py

@@ -1,391 +1,85 @@
+from __future__ import annotations
+
 import time
-import uuid
-import warnings
-from collections import OrderedDict
 from typing import Any
 
 import gradio as gr
 import spaces
 
-from src.constants import PARAKEET_V3
-from src.diarization_service import run_pyannote_diarization_request
-from src.merge_service import merge_parakeet_pyannote_outputs
-from src.models.parakeet_model import preload_parakeet_model, run_parakeet
-from src.models.pyannote_community_model import preload_pyannote_pipeline, run_pyannote_community_chunk
-from src.transcription_service import run_parakeet_transcription_request
-
-# Suppress a known deprecation warning emitted by a transitive dependency in spaces.
-warnings.filterwarnings(
-    "ignore",
-    message=r"`torch\.distributed\.reduce_op` is deprecated, please use `torch\.distributed\.ReduceOp` instead",
-    category=FutureWarning,
-)
+from src.parakeet.config import PARAKEET_MODEL_LABEL
+from src.parakeet.model import preload_parakeet_model, run_parakeet_inference
+from src.parakeet.service import run_transcribe_parakeet_request
+from src.runtime.errors import format_exception_for_client
+from src.runtime.logging import configure_logging, get_logger, log_event
 
-_PRELOAD_ERRORS: dict[str, str] = {}
-_DEBUG_RUNS: "OrderedDict[str, dict[str, Any]]" = OrderedDict()
-_MAX_DEBUG_RUNS = 10
-_LAST_DEBUG_RUN_ID: str | None = None
+configure_logging()
+logger = get_logger(__name__)
 
-RUN_COMPLETE_PIPELINE_OUTPUT_SCHEMA: dict[str, Any] = {
-    "type": "object",
-    "description": "Merged transcript output from Parakeet (word timestamps) and Pyannote diarization.",
-    "properties": {
-        "summary": {
-            "type": "object",
-            "properties": {
-                "diarization_key_used": {"type": "string", "example": "exclusive_speaker_diarization"},
-                "parakeet_word_count": {"type": "integer"},
-                "pyannote_segment_count": {"type": "integer"},
-                "turn_count": {"type": "integer"},
-                "assigned_word_count": {"type": "integer"},
-                "unassigned_word_count": {"type": "integer"},
-            },
-            "required": [
-                "diarization_key_used",
-                "parakeet_word_count",
-                "pyannote_segment_count",
-                "turn_count",
-                "assigned_word_count",
-                "unassigned_word_count",
-            ],
-        },
-        "turns": {
-            "type": "array",
-            "items": {
-                "type": "object",
-                "properties": {
-                    "speaker": {"type": "string", "example": "SPEAKER_02"},
-                    "start": {"type": "number", "example": 40.72},
-                    "end": {"type": "number", "example": 514.0},
-                    "text": {"type": "string"},
-                },
-                "required": ["speaker", "start", "end", "text"],
-            },
-        },
-        "transcript_text": {"type": "string"},
-    },
-    "required": ["summary", "turns", "transcript_text"],
-}
-
-RUN_COMPLETE_PIPELINE_OUTPUT_EXAMPLE: dict[str, Any] = {
-    "summary": {
-        "diarization_key_used": "exclusive_speaker_diarization",
-        "parakeet_word_count": 1234,
-        "pyannote_segment_count": 42,
-        "turn_count": 39,
-        "assigned_word_count": 1219,
-        "unassigned_word_count": 15,
-    },
-    "turns": [
-        {
-            "speaker": "SPEAKER_00",
-            "start": 0.0,
-            "end": 12.34,
-            "text": "Good morning and welcome to the earnings call.",
-        },
-        {
-            "speaker": "SPEAKER_01",
-            "start": 12.34,
-            "end": 19.02,
-            "text": "Thank you. Let us begin with quarterly highlights.",
-        },
-    ],
-    "transcript_text": "[0.00 - 12.34] SPEAKER_00: Good morning ...",
-}
-
-RUN_COMPLETE_PIPELINE_INPUT_SCHEMA: dict[str, Any] = {
-    "type": "object",
-    "properties": {
-        "audio_file": {"type": "file", "description": "Audio file upload"},
-        "huggingface_token": {"type": "string", "description": "HF access token for pyannote model"},
-    },
-    "required": ["audio_file", "huggingface_token"],
-}
-
-
-def _preload_model(model_label: str, preload_fn) -> None:
-    try:
-        preload_fn()
-    except Exception as exc:
-        _PRELOAD_ERRORS[model_label] = str(exc)
-
-
-def _raise_preload_error_if_any(model_label: str) -> None:
-    message = _PRELOAD_ERRORS.get(model_label)
-    if message:
-        raise gr.Error(
-            f"Model preload failed for {model_label}. "
-            "Check startup logs and dependencies. "
-            f"Details: {message}"
-        )
-
-
-def _store_debug_payload(payload: dict[str, Any]) -> str:
-    global _LAST_DEBUG_RUN_ID
-    run_id = str(uuid.uuid4())
-    _DEBUG_RUNS[run_id] = payload
-    _LAST_DEBUG_RUN_ID = run_id
-    while len(_DEBUG_RUNS) > _MAX_DEBUG_RUNS:
-        _DEBUG_RUNS.popitem(last=False)
-    return run_id
-
-
-# Global setup (outside @spaces.GPU) so setup cost is not charged to ZeroGPU inference window.
-_preload_model(PARAKEET_V3, preload_parakeet_model)
-# Pyannote preload is best-effort at startup because token is provided per request.
-preload_pyannote_pipeline(strict=False)
+_STARTUP_PRELOAD = preload_parakeet_model(strict=False)
+log_event(logger, "startup.parakeet_preload", status=_STARTUP_PRELOAD)
 
 
 @spaces.GPU(duration=120)
-def _gpu_infer_parakeet(audio_file: str, duration_seconds: float | None):
-    gpu_start = time.perf_counter()
-    result = run_parakeet(
+def _gpu_transcribe_parakeet(
+    audio_file: str,
+    duration_seconds: float | None,
+    model_options: dict[str, Any],
+) -> dict[str, Any]:
+    gpu_started_at = time.perf_counter()
+    result = run_parakeet_inference(
         audio_file=audio_file,
-        language=None,
-        model_options={},
         duration_seconds=duration_seconds,
-    )
-    gpu_end = time.perf_counter()
-    return {
-        "raw_output": result["raw_output"],
-        "zerogpu_timing": {
-            "gpu_window_seconds": round(gpu_end - gpu_start, 4),
-            **result.get("timing", {}),
-        },
-    }
-
-
-@spaces.GPU(duration=120)
-def _gpu_infer_pyannote_chunk(audio_file: str, model_options: dict[str, Any]):
-    gpu_start = time.perf_counter()
-    result = run_pyannote_community_chunk(
-        audio_file=audio_file,
         model_options=model_options,
     )
-    gpu_end = time.perf_counter()
     return {
         "raw_output": result["raw_output"],
         "zerogpu_timing": {
-            "gpu_window_seconds": round(gpu_end - gpu_start, 4),
-            **result.get("timing", {}),
-        },
-    }
-
-
-def transcribe(audio_file: str):
-    _raise_preload_error_if_any(PARAKEET_V3)
-    return run_parakeet_transcription_request(
-        audio_file=audio_file,
-        gpu_runner=_gpu_infer_parakeet,
-    )
-
-
-def diarize(audio_file: str, huggingface_token: str):
-    return run_pyannote_diarization_request(
-        audio_file=audio_file,
-        huggingface_token=huggingface_token,
-        preload_runner=preload_pyannote_pipeline,
-        gpu_chunk_runner=_gpu_infer_pyannote_chunk,
-    )
-
-
-def run_complete_pipeline(
-    audio_file: str,
-    huggingface_token: str,
-):
-    _raise_preload_error_if_any(PARAKEET_V3)
-
-    started_at = time.perf_counter()
-
-    # 1) Parakeet transcription on ZeroGPU.
-    parakeet_response = run_parakeet_transcription_request(
-        audio_file=audio_file,
-        gpu_runner=_gpu_infer_parakeet,
-    )
-
-    # 2) Pyannote diarization on ZeroGPU (chunked only when needed).
-    pyannote_response = run_pyannote_diarization_request(
-        audio_file=audio_file,
-        huggingface_token=huggingface_token,
-        preload_runner=preload_pyannote_pipeline,
-        gpu_chunk_runner=_gpu_infer_pyannote_chunk,
-    )
-
-    # 3) CPU-side postprocessing outside ZeroGPU.
-    merged_transcript = merge_parakeet_pyannote_outputs(
-        parakeet_response=parakeet_response,
-        pyannote_response=pyannote_response,
-        diarization_key="exclusive_speaker_diarization",
-    )
-
-    total_gpu_window_seconds = float(parakeet_response["zerogpu_timing"].get("gpu_window_seconds", 0.0)) + float(
-        pyannote_response.get("zerogpu_timing", {}).get("gpu_window_seconds", 0.0)
-    )
-    total_inference_seconds = float(parakeet_response["zerogpu_timing"].get("inference_seconds", 0.0)) + float(
-        pyannote_response.get("zerogpu_timing", {}).get("inference_seconds", 0.0)
-    )
-
-    finished_at = time.perf_counter()
-    debug_payload = {
-        "pipeline_timing": {
-            "total_wall_clock_seconds": round(finished_at - started_at, 4),
-            "zerogpu_gpu_window_seconds_total": round(total_gpu_window_seconds, 4),
-            "zerogpu_inference_seconds_total": round(total_inference_seconds, 4),
-        },
-        "inputs": {
-            "audio_file": str(audio_file),
-            "huggingface_token_provided": bool(huggingface_token),
+            "gpu_window_seconds": round(time.perf_counter() - gpu_started_at, 4),
+            **result["timing"],
         },
-        "parakeet_response": parakeet_response,
-        "pyannote_response": pyannote_response,
-        "merged_transcript": merged_transcript,
     }
-    _store_debug_payload(debug_payload)
 
-    # Return merged transcript JSON (OpenAI cleanup is intentionally local/off-space).
-    return merged_transcript
 
-
-def get_debug_output(run_id: str | None):
-    if run_id and run_id.strip():
-        payload = _DEBUG_RUNS.get(run_id.strip())
-        if payload is None:
-            raise gr.Error(f"Unknown run_id: {run_id}")
-        return {"run_id": run_id.strip(), "debug": payload}
-
-    if _LAST_DEBUG_RUN_ID is None:
-        raise gr.Error("No debug payload available yet. Run /run_complete_pipeline first.")
-    return {"run_id": _LAST_DEBUG_RUN_ID, "debug": _DEBUG_RUNS[_LAST_DEBUG_RUN_ID]}
-
-
-def get_run_complete_pipeline_schema() -> dict[str, Any]:
-    return {
-        "api_name": "/run_complete_pipeline",
-        "input_schema": RUN_COMPLETE_PIPELINE_INPUT_SCHEMA,
-        "output_schema": RUN_COMPLETE_PIPELINE_OUTPUT_SCHEMA,
-        "output_example": RUN_COMPLETE_PIPELINE_OUTPUT_EXAMPLE,
-        "notes": [
-            "Use /get_debug_output to fetch raw model payloads and timing.",
-            "The production route returns only merged transcript JSON.",
-        ],
-    }
+def transcribe_parakeet(audio_file: str, model_options_json: str | None) -> dict[str, Any]:
+    try:
+        return run_transcribe_parakeet_request(
+            audio_file=audio_file,
+            model_options_json=model_options_json,
+            gpu_runner=_gpu_transcribe_parakeet,
+        )
+    except Exception as exc:
+        logger.exception("Parakeet request failed")
+        raise gr.Error(format_exception_for_client(exc)) from exc
 
 
-with gr.Blocks(title="Parakeet + Pyannote Pipeline") as demo:
+with gr.Blocks(title="Phase 1 - Parakeet only") as demo:
     gr.Markdown(
-        "# End-to-end transcript pipeline\n"
-        "Runs Parakeet transcription, Pyannote diarization, then merges into a combined transcript JSON.\n\n"
-        "Need stage isolation? Call `/transcribe` or `/diarize`."
-    )
-
-    audio_file = gr.Audio(
-        sources=["upload"],
-        type="filepath",
-        label="Audio file",
-    )
-    transcribe_btn = gr.Button("Transcribe only")
-    transcribe_output = gr.JSON(label="Parakeet transcript JSON")
-    transcribe_btn.click(
-        fn=transcribe,
-        inputs=[audio_file],
-        outputs=transcribe_output,
-        api_name="transcribe",
-        api_description=(
-            "Run Parakeet only and return the same transcription payload used inside /run_complete_pipeline.\n"
-            "Response shape:\n"
-            "{\n"
-            '  "model": "NVIDIA Parakeet v3",\n'
-            '  "task": "transcribe",\n'
-            '  "audio_file": str,\n'
-            '  "postprocess_prompt": null,\n'
-            '  "model_options": {},\n'
-            '  "zerogpu_timing": {"gpu_window_seconds": float, "inference_seconds": float},\n'
-            '  "raw_output": dict,\n'
-            '  "timestamp_granularity": "word"\n'
-            "}"
-        ),
+        "# Phase 1 rebuild: Parakeet-only transcription\n\n"
+        "The old implementation has been archived. This Space now exposes only the Phase 1 "
+        "Parakeet transcription API while ZeroGPU usage is benchmarked and stabilized."
     )
 
-    huggingface_token = gr.Textbox(
-        label="HuggingFace token",
-        type="password",
-    )
-    diarize_btn = gr.Button("Diarize only")
-    diarize_output = gr.JSON(label="Pyannote diarization JSON")
-    diarize_btn.click(
-        fn=diarize,
-        inputs=[audio_file, huggingface_token],
-        outputs=diarize_output,
-        api_name="diarize",
-        api_description=(
-            "Run Pyannote only and return the same diarization payload used inside /run_complete_pipeline.\n"
-            "Requires huggingface_token for pyannote/speaker-diarization-community-1.\n"
-            "Response shape:\n"
-            "{\n"
-            '  "model": "pyannote/speaker-diarization-community-1",\n'
-            '  "audio_file": str,\n'
-            '  "duration_seconds": float | null,\n'
-            '  "model_options": dict,\n'
-            '  "long_audio": {"chunk_count": int, "chunk_plan": list, "chunked": bool},\n'
-            '  "zerogpu_timing": {"gpu_window_seconds": float, "inference_seconds": float},\n'
-            '  "raw_output": {"chunk_results": list, "stitched": dict}\n'
-            "}"
-        ),
+    audio_file = gr.Audio(sources=["upload"], type="filepath", label="Audio file")
+    model_options_json = gr.Textbox(
+        label="model_options_json",
+        lines=8,
+        value="{}",
+        placeholder='{"compute_dtype": "bfloat16"}',
     )
+    run_button = gr.Button("Run /transcribe_parakeet")
+    output = gr.JSON(label="Parakeet transcription JSON")
 
-    run_btn = gr.Button("Run full pipeline")
-    output = gr.JSON(label="Combined transcript JSON")
-
-    run_btn.click(
-        fn=run_complete_pipeline,
-        inputs=[audio_file, huggingface_token],
+    run_button.click(
+        fn=transcribe_parakeet,
+        inputs=[audio_file, model_options_json],
         outputs=output,
-        api_name="run_complete_pipeline",
-        api_description=(
-            "Run Parakeet + Pyannote and return merged transcript JSON.\n"
-            "Response shape:\n"
-            "{\n"
-            '  "summary": {\n'
-            '    "diarization_key_used": str,\n'
-            '    "parakeet_word_count": int,\n'
-            '    "pyannote_segment_count": int,\n'
-            '    "turn_count": int,\n'
-            '    "assigned_word_count": int,\n'
-            '    "unassigned_word_count": int\n'
-            "  },\n"
-            '  "turns": [{"speaker": str, "start": float, "end": float, "text": str}],\n'
-            '  "transcript_text": str\n'
-            "}\n"
-            "For full machine-readable schema + example, call /get_run_complete_pipeline_schema."
-        ),
-    )
-
-    with gr.Row():
-        debug_run_id = gr.Textbox(label="Debug run_id (optional)")
-        debug_btn = gr.Button("Get debug output")
-    debug_output = gr.JSON(label="Debug output (raw + benchmark)")
-
-    debug_btn.click(
-        fn=get_debug_output,
-        inputs=[debug_run_id],
-        outputs=debug_output,
-        api_name="get_debug_output",
+        api_name="transcribe_parakeet",
         api_description=(
-            "Return latest (or selected) debug payload including raw Parakeet/Pyannote outputs "
-            "and aggregated pipeline timing."
+            f"Run {PARAKEET_MODEL_LABEL} only. "
+            "This Phase 1 route performs CPU-side duration probe/preload/postprocessing and keeps the "
+            "decorated ZeroGPU window narrowly scoped to model inference."
         ),
     )
 
-    with gr.Row(visible=False):
-        schema_btn = gr.Button("get_run_complete_pipeline_schema")
-    schema_output = gr.JSON(label="run_complete_pipeline schema", visible=False)
-    schema_btn.click(
-        fn=get_run_complete_pipeline_schema,
-        inputs=None,
-        outputs=schema_output,
-        api_name="get_run_complete_pipeline_schema",
-        api_description="Return input/output schema contract for /run_complete_pipeline.",
-    )
-
 
 demo.queue(default_concurrency_limit=1).launch(ssr_mode=False, theme=gr.themes.Ocean())

archive/legacy_pre_rebuild_20260407_133100/MANIFEST.md

@@ -0,0 +1,23 @@
+# Legacy archive: pre-rebuild implementation
+
+This directory contains the implementation that existed before the phased rebuild began.
+
+Archived to prevent old code from polluting the Phase 1 rebuild context.
+
+## What was moved here
+- Previous `app.py`
+- Previous `README.md`
+- Previous `requirements.txt` and `packages.txt`
+- Entire previous `src/` tree
+- Entire previous `tests/` tree
+- Benchmark/output artifacts and sample audio assets
+- Local helper scripts tied to the old combined pipeline
+
+## What was intentionally kept at repo root
+- `.git/`
+- `.gitattributes`
+- `.gitignore`
+- `plan/`
+
+## Rebuild rule
+The active implementation at repo root should now reflect only the new phased rebuild.

archive/legacy_pre_rebuild_20260407_133100/README.md

@@ -0,0 +1,173 @@
+---
+title: Transcribe Diarize
+emoji: ⚡
+colorFrom: indigo
+colorTo: indigo
+sdk: gradio
+sdk_version: 6.6.0
+python_version: "3.12"
+app_file: app.py
+pinned: false
+license: mit
+short_description: End-to-end Parakeet + Pyannote transcript pipeline
+---
+
+This Space is optimized for API usage on ZeroGPU.
+
+## Pipeline
+Single production pipeline:
+1. Parakeet transcription (word timestamps) on ZeroGPU
+2. Pyannote Community-1 diarization on ZeroGPU
+3. Merge transcript + diarization (CPU)
+
+Model setup is global/outside `@spaces.GPU` so setup time is not billed to ZeroGPU windows.
+
+## API routes
+- `/transcribe`
+- `/diarize`
+- `/run_complete_pipeline`
+- `/get_debug_output`
+- `/get_run_complete_pipeline_schema`
+
+## `/transcribe` inputs
+- `audio_file` (file path from Gradio client upload)
+
+Returns: Parakeet-only transcription JSON, exactly the payload used inside the full pipeline:
+
+```json
+{
+  "model": "NVIDIA Parakeet v3",
+  "task": "transcribe",
+  "audio_file": "/tmp/gradio/example.wav",
+  "postprocess_prompt": null,
+  "model_options": {},
+  "zerogpu_timing": {
+    "gpu_window_seconds": 3.12,
+    "inference_seconds": 2.87
+  },
+  "raw_output": {
+    "output": {},
+    "timestamp_hint": "word timestamps available in output.timestamp['word'] when provided by NeMo",
+    "language_hint": "auto",
+    "long_audio_settings": {}
+  },
+  "timestamp_granularity": "word"
+}
+```
+
+## `/diarize` inputs
+- `audio_file` (file path from Gradio client upload)
+- `huggingface_token`
+
+Returns: Pyannote-only diarization JSON, exactly the payload used inside the full pipeline:
+
+```json
+{
+  "model": "pyannote/speaker-diarization-community-1",
+  "audio_file": "/tmp/gradio/example.wav",
+  "duration_seconds": 300.0,
+  "model_options": {
+    "hf_token": "***redacted***",
+    "long_audio_chunk_threshold_s": 7200,
+    "chunk_duration_s": 7200,
+    "chunk_overlap_s": 0
+  },
+  "long_audio": {
+    "chunk_count": 1,
+    "chunk_plan": [{"index": 0, "start": 0.0, "end": 300.0, "duration": 300.0}],
+    "chunked": false
+  },
+  "zerogpu_timing": {
+    "gpu_window_seconds": 8.45,
+    "inference_seconds": 8.11
+  },
+  "raw_output": {
+    "chunk_results": [],
+    "stitched": {
+      "speaker_diarization": [],
+      "exclusive_speaker_diarization": []
+    }
+  }
+}
+```
+
+## `/run_complete_pipeline` inputs
+- `audio_file` (file path from Gradio client upload)
+- `huggingface_token`
+
+Returns: merged transcript JSON only, with this structure:
+
+```json
+{
+  "summary": {
+    "diarization_key_used": "exclusive_speaker_diarization",
+    "parakeet_word_count": 1234,
+    "pyannote_segment_count": 42,
+    "turn_count": 39,
+    "assigned_word_count": 1219,
+    "unassigned_word_count": 15
+  },
+  "turns": [
+    {
+      "speaker": "SPEAKER_00",
+      "start": 0.0,
+      "end": 12.34,
+      "text": "Good morning and welcome to the earnings call."
+    }
+  ],
+  "transcript_text": "[0.00 - 12.34] SPEAKER_00: Good morning and welcome to the earnings call."
+}
+```
+
+## `/get_debug_output` inputs
+- `run_id` (optional)
+
+Returns: raw/debug payload for the latest run (or specific run if provided), including:
+- Parakeet raw output + ZeroGPU timing
+- Pyannote raw output + ZeroGPU timing
+- merged transcript payload
+- aggregated timing
+
+## `/get_run_complete_pipeline_schema` inputs
+- none
+
+Returns a machine-readable contract for `/run_complete_pipeline`:
+- `input_schema`
+- `output_schema`
+- `output_example`
+
+## IPython example
+```python
+from gradio_client import Client, handle_file
+
+SPACE = "username/space-name"
+AUDIO_FILE = "Q3-FY26_5min.mp3"
+
+client = Client(SPACE)
+
+# Run Parakeet only
+parakeet_response = client.predict(
+    audio_file=handle_file(AUDIO_FILE),
+    api_name="/transcribe",
+)
+
+# Run Pyannote only
+pyannote_response = client.predict(
+    audio_file=handle_file(AUDIO_FILE),
+    huggingface_token="hf_xxx",
+    api_name="/diarize",
+)
+
+# Run end-to-end pipeline (returns merged transcript)
+merged_transcript = client.predict(
+    audio_file=handle_file(AUDIO_FILE),
+    huggingface_token="hf_xxx",
+    api_name="/run_complete_pipeline",
+)
+
+# Fetch latest debug payload
+debug_payload = client.predict(
+    run_id=None,  # or a specific run_id returned by prior debug call
+    api_name="/get_debug_output",
+)
+```

archive/legacy_pre_rebuild_20260407_133100/app.py

@@ -0,0 +1,391 @@
+import time
+import uuid
+import warnings
+from collections import OrderedDict
+from typing import Any
+
+import gradio as gr
+import spaces
+
+from src.constants import PARAKEET_V3
+from src.diarization_service import run_pyannote_diarization_request
+from src.merge_service import merge_parakeet_pyannote_outputs
+from src.models.parakeet_model import preload_parakeet_model, run_parakeet
+from src.models.pyannote_community_model import preload_pyannote_pipeline, run_pyannote_community_chunk
+from src.transcription_service import run_parakeet_transcription_request
+
+# Suppress a known deprecation warning emitted by a transitive dependency in spaces.
+warnings.filterwarnings(
+    "ignore",
+    message=r"`torch\.distributed\.reduce_op` is deprecated, please use `torch\.distributed\.ReduceOp` instead",
+    category=FutureWarning,
+)
+
+_PRELOAD_ERRORS: dict[str, str] = {}
+_DEBUG_RUNS: "OrderedDict[str, dict[str, Any]]" = OrderedDict()
+_MAX_DEBUG_RUNS = 10
+_LAST_DEBUG_RUN_ID: str | None = None
+
+RUN_COMPLETE_PIPELINE_OUTPUT_SCHEMA: dict[str, Any] = {
+    "type": "object",
+    "description": "Merged transcript output from Parakeet (word timestamps) and Pyannote diarization.",
+    "properties": {
+        "summary": {
+            "type": "object",
+            "properties": {
+                "diarization_key_used": {"type": "string", "example": "exclusive_speaker_diarization"},
+                "parakeet_word_count": {"type": "integer"},
+                "pyannote_segment_count": {"type": "integer"},
+                "turn_count": {"type": "integer"},
+                "assigned_word_count": {"type": "integer"},
+                "unassigned_word_count": {"type": "integer"},
+            },
+            "required": [
+                "diarization_key_used",
+                "parakeet_word_count",
+                "pyannote_segment_count",
+                "turn_count",
+                "assigned_word_count",
+                "unassigned_word_count",
+            ],
+        },
+        "turns": {
+            "type": "array",
+            "items": {
+                "type": "object",
+                "properties": {
+                    "speaker": {"type": "string", "example": "SPEAKER_02"},
+                    "start": {"type": "number", "example": 40.72},
+                    "end": {"type": "number", "example": 514.0},
+                    "text": {"type": "string"},
+                },
+                "required": ["speaker", "start", "end", "text"],
+            },
+        },
+        "transcript_text": {"type": "string"},
+    },
+    "required": ["summary", "turns", "transcript_text"],
+}
+
+RUN_COMPLETE_PIPELINE_OUTPUT_EXAMPLE: dict[str, Any] = {
+    "summary": {
+        "diarization_key_used": "exclusive_speaker_diarization",
+        "parakeet_word_count": 1234,
+        "pyannote_segment_count": 42,
+        "turn_count": 39,
+        "assigned_word_count": 1219,
+        "unassigned_word_count": 15,
+    },
+    "turns": [
+        {
+            "speaker": "SPEAKER_00",
+            "start": 0.0,
+            "end": 12.34,
+            "text": "Good morning and welcome to the earnings call.",
+        },
+        {
+            "speaker": "SPEAKER_01",
+            "start": 12.34,
+            "end": 19.02,
+            "text": "Thank you. Let us begin with quarterly highlights.",
+        },
+    ],
+    "transcript_text": "[0.00 - 12.34] SPEAKER_00: Good morning ...",
+}
+
+RUN_COMPLETE_PIPELINE_INPUT_SCHEMA: dict[str, Any] = {
+    "type": "object",
+    "properties": {
+        "audio_file": {"type": "file", "description": "Audio file upload"},
+        "huggingface_token": {"type": "string", "description": "HF access token for pyannote model"},
+    },
+    "required": ["audio_file", "huggingface_token"],
+}
+
+
+def _preload_model(model_label: str, preload_fn) -> None:
+    try:
+        preload_fn()
+    except Exception as exc:
+        _PRELOAD_ERRORS[model_label] = str(exc)
+
+
+def _raise_preload_error_if_any(model_label: str) -> None:
+    message = _PRELOAD_ERRORS.get(model_label)
+    if message:
+        raise gr.Error(
+            f"Model preload failed for {model_label}. "
+            "Check startup logs and dependencies. "
+            f"Details: {message}"
+        )
+
+
+def _store_debug_payload(payload: dict[str, Any]) -> str:
+    global _LAST_DEBUG_RUN_ID
+    run_id = str(uuid.uuid4())
+    _DEBUG_RUNS[run_id] = payload
+    _LAST_DEBUG_RUN_ID = run_id
+    while len(_DEBUG_RUNS) > _MAX_DEBUG_RUNS:
+        _DEBUG_RUNS.popitem(last=False)
+    return run_id
+
+
+# Global setup (outside @spaces.GPU) so setup cost is not charged to ZeroGPU inference window.
+_preload_model(PARAKEET_V3, preload_parakeet_model)
+# Pyannote preload is best-effort at startup because token is provided per request.
+preload_pyannote_pipeline(strict=False)
+
+
+@spaces.GPU(duration=120)
+def _gpu_infer_parakeet(audio_file: str, duration_seconds: float | None):
+    gpu_start = time.perf_counter()
+    result = run_parakeet(
+        audio_file=audio_file,
+        language=None,
+        model_options={},
+        duration_seconds=duration_seconds,
+    )
+    gpu_end = time.perf_counter()
+    return {
+        "raw_output": result["raw_output"],
+        "zerogpu_timing": {
+            "gpu_window_seconds": round(gpu_end - gpu_start, 4),
+            **result.get("timing", {}),
+        },
+    }
+
+
+@spaces.GPU(duration=120)
+def _gpu_infer_pyannote_chunk(audio_file: str, model_options: dict[str, Any]):
+    gpu_start = time.perf_counter()
+    result = run_pyannote_community_chunk(
+        audio_file=audio_file,
+        model_options=model_options,
+    )
+    gpu_end = time.perf_counter()
+    return {
+        "raw_output": result["raw_output"],
+        "zerogpu_timing": {
+            "gpu_window_seconds": round(gpu_end - gpu_start, 4),
+            **result.get("timing", {}),
+        },
+    }
+
+
+def transcribe(audio_file: str):
+    _raise_preload_error_if_any(PARAKEET_V3)
+    return run_parakeet_transcription_request(
+        audio_file=audio_file,
+        gpu_runner=_gpu_infer_parakeet,
+    )
+
+
+def diarize(audio_file: str, huggingface_token: str):
+    return run_pyannote_diarization_request(
+        audio_file=audio_file,
+        huggingface_token=huggingface_token,
+        preload_runner=preload_pyannote_pipeline,
+        gpu_chunk_runner=_gpu_infer_pyannote_chunk,
+    )
+
+
+def run_complete_pipeline(
+    audio_file: str,
+    huggingface_token: str,
+):
+    _raise_preload_error_if_any(PARAKEET_V3)
+
+    started_at = time.perf_counter()
+
+    # 1) Parakeet transcription on ZeroGPU.
+    parakeet_response = run_parakeet_transcription_request(
+        audio_file=audio_file,
+        gpu_runner=_gpu_infer_parakeet,
+    )
+
+    # 2) Pyannote diarization on ZeroGPU (chunked only when needed).
+    pyannote_response = run_pyannote_diarization_request(
+        audio_file=audio_file,
+        huggingface_token=huggingface_token,
+        preload_runner=preload_pyannote_pipeline,
+        gpu_chunk_runner=_gpu_infer_pyannote_chunk,
+    )
+
+    # 3) CPU-side postprocessing outside ZeroGPU.
+    merged_transcript = merge_parakeet_pyannote_outputs(
+        parakeet_response=parakeet_response,
+        pyannote_response=pyannote_response,
+        diarization_key="exclusive_speaker_diarization",
+    )
+
+    total_gpu_window_seconds = float(parakeet_response["zerogpu_timing"].get("gpu_window_seconds", 0.0)) + float(
+        pyannote_response.get("zerogpu_timing", {}).get("gpu_window_seconds", 0.0)
+    )
+    total_inference_seconds = float(parakeet_response["zerogpu_timing"].get("inference_seconds", 0.0)) + float(
+        pyannote_response.get("zerogpu_timing", {}).get("inference_seconds", 0.0)
+    )
+
+    finished_at = time.perf_counter()
+    debug_payload = {
+        "pipeline_timing": {
+            "total_wall_clock_seconds": round(finished_at - started_at, 4),
+            "zerogpu_gpu_window_seconds_total": round(total_gpu_window_seconds, 4),
+            "zerogpu_inference_seconds_total": round(total_inference_seconds, 4),
+        },
+        "inputs": {
+            "audio_file": str(audio_file),
+            "huggingface_token_provided": bool(huggingface_token),
+        },
+        "parakeet_response": parakeet_response,
+        "pyannote_response": pyannote_response,
+        "merged_transcript": merged_transcript,
+    }
+    _store_debug_payload(debug_payload)
+
+    # Return merged transcript JSON (OpenAI cleanup is intentionally local/off-space).
+    return merged_transcript
+
+
+def get_debug_output(run_id: str | None):
+    if run_id and run_id.strip():
+        payload = _DEBUG_RUNS.get(run_id.strip())
+        if payload is None:
+            raise gr.Error(f"Unknown run_id: {run_id}")
+        return {"run_id": run_id.strip(), "debug": payload}
+
+    if _LAST_DEBUG_RUN_ID is None:
+        raise gr.Error("No debug payload available yet. Run /run_complete_pipeline first.")
+    return {"run_id": _LAST_DEBUG_RUN_ID, "debug": _DEBUG_RUNS[_LAST_DEBUG_RUN_ID]}
+
+
+def get_run_complete_pipeline_schema() -> dict[str, Any]:
+    return {
+        "api_name": "/run_complete_pipeline",
+        "input_schema": RUN_COMPLETE_PIPELINE_INPUT_SCHEMA,
+        "output_schema": RUN_COMPLETE_PIPELINE_OUTPUT_SCHEMA,
+        "output_example": RUN_COMPLETE_PIPELINE_OUTPUT_EXAMPLE,
+        "notes": [
+            "Use /get_debug_output to fetch raw model payloads and timing.",
+            "The production route returns only merged transcript JSON.",
+        ],
+    }
+
+
+with gr.Blocks(title="Parakeet + Pyannote Pipeline") as demo:
+    gr.Markdown(
+        "# End-to-end transcript pipeline\n"
+        "Runs Parakeet transcription, Pyannote diarization, then merges into a combined transcript JSON.\n\n"
+        "Need stage isolation? Call `/transcribe` or `/diarize`."
+    )
+
+    audio_file = gr.Audio(
+        sources=["upload"],
+        type="filepath",
+        label="Audio file",
+    )
+    transcribe_btn = gr.Button("Transcribe only")
+    transcribe_output = gr.JSON(label="Parakeet transcript JSON")
+    transcribe_btn.click(
+        fn=transcribe,
+        inputs=[audio_file],
+        outputs=transcribe_output,
+        api_name="transcribe",
+        api_description=(
+            "Run Parakeet only and return the same transcription payload used inside /run_complete_pipeline.\n"
+            "Response shape:\n"
+            "{\n"
+            '  "model": "NVIDIA Parakeet v3",\n'
+            '  "task": "transcribe",\n'
+            '  "audio_file": str,\n'
+            '  "postprocess_prompt": null,\n'
+            '  "model_options": {},\n'
+            '  "zerogpu_timing": {"gpu_window_seconds": float, "inference_seconds": float},\n'
+            '  "raw_output": dict,\n'
+            '  "timestamp_granularity": "word"\n'
+            "}"
+        ),
+    )
+
+    huggingface_token = gr.Textbox(
+        label="HuggingFace token",
+        type="password",
+    )
+    diarize_btn = gr.Button("Diarize only")
+    diarize_output = gr.JSON(label="Pyannote diarization JSON")
+    diarize_btn.click(
+        fn=diarize,
+        inputs=[audio_file, huggingface_token],
+        outputs=diarize_output,
+        api_name="diarize",
+        api_description=(
+            "Run Pyannote only and return the same diarization payload used inside /run_complete_pipeline.\n"
+            "Requires huggingface_token for pyannote/speaker-diarization-community-1.\n"
+            "Response shape:\n"
+            "{\n"
+            '  "model": "pyannote/speaker-diarization-community-1",\n'
+            '  "audio_file": str,\n'
+            '  "duration_seconds": float | null,\n'
+            '  "model_options": dict,\n'
+            '  "long_audio": {"chunk_count": int, "chunk_plan": list, "chunked": bool},\n'
+            '  "zerogpu_timing": {"gpu_window_seconds": float, "inference_seconds": float},\n'
+            '  "raw_output": {"chunk_results": list, "stitched": dict}\n'
+            "}"
+        ),
+    )
+
+    run_btn = gr.Button("Run full pipeline")
+    output = gr.JSON(label="Combined transcript JSON")
+
+    run_btn.click(
+        fn=run_complete_pipeline,
+        inputs=[audio_file, huggingface_token],
+        outputs=output,
+        api_name="run_complete_pipeline",
+        api_description=(
+            "Run Parakeet + Pyannote and return merged transcript JSON.\n"
+            "Response shape:\n"
+            "{\n"
+            '  "summary": {\n'
+            '    "diarization_key_used": str,\n'
+            '    "parakeet_word_count": int,\n'
+            '    "pyannote_segment_count": int,\n'
+            '    "turn_count": int,\n'
+            '    "assigned_word_count": int,\n'
+            '    "unassigned_word_count": int\n'
+            "  },\n"
+            '  "turns": [{"speaker": str, "start": float, "end": float, "text": str}],\n'
+            '  "transcript_text": str\n'
+            "}\n"
+            "For full machine-readable schema + example, call /get_run_complete_pipeline_schema."
+        ),
+    )
+
+    with gr.Row():
+        debug_run_id = gr.Textbox(label="Debug run_id (optional)")
+        debug_btn = gr.Button("Get debug output")
+    debug_output = gr.JSON(label="Debug output (raw + benchmark)")
+
+    debug_btn.click(
+        fn=get_debug_output,
+        inputs=[debug_run_id],
+        outputs=debug_output,
+        api_name="get_debug_output",
+        api_description=(
+            "Return latest (or selected) debug payload including raw Parakeet/Pyannote outputs "
+            "and aggregated pipeline timing."
+        ),
+    )
+
+    with gr.Row(visible=False):
+        schema_btn = gr.Button("get_run_complete_pipeline_schema")
+    schema_output = gr.JSON(label="run_complete_pipeline schema", visible=False)
+    schema_btn.click(
+        fn=get_run_complete_pipeline_schema,
+        inputs=None,
+        outputs=schema_output,
+        api_name="get_run_complete_pipeline_schema",
+        api_description="Return input/output schema contract for /run_complete_pipeline.",
+    )
+
+
+demo.queue(default_concurrency_limit=1).launch(ssr_mode=False, theme=gr.themes.Ocean())

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260221_183454/benchmark_stats.json

@@ -0,0 +1,69 @@
+{
+  "space": "Ratnesh-dev/transcribe-diarize",
+  "audio_file": "Q3-FY26_5min.mp3",
+  "task": "transcribe",
+  "language": null,
+  "models": [
+    "Whisper Large V3",
+    "Whisper Large V3 Turbo",
+    "Whisper.cpp (large)",
+    "Whisper faster (large)",
+    "NVIDIA Parakeet v3"
+  ],
+  "benchmark_timing": {
+    "total_client_wall_clock_seconds": 158.1297
+  },
+  "leaderboard_by_gpu_window_seconds": [
+    {
+      "model": "NVIDIA Parakeet v3",
+      "api_name": "/transcribe_parakeet_v3",
+      "gpu_window_seconds": 25.6806,
+      "inference_seconds": 1.6436,
+      "client_wall_clock_seconds": 35.6289
+    },
+    {
+      "model": "Whisper faster (large)",
+      "api_name": "/transcribe_whisper_faster_large",
+      "gpu_window_seconds": 26.2386,
+      "inference_seconds": 23.6107,
+      "client_wall_clock_seconds": 32.4749
+    }
+  ],
+  "summary": [
+    {
+      "model": "Whisper Large V3",
+      "status": "error",
+      "client_wall_clock_seconds": 32.93,
+      "gpu_window_seconds": null,
+      "inference_seconds": null
+    },
+    {
+      "model": "Whisper Large V3 Turbo",
+      "status": "error",
+      "client_wall_clock_seconds": 49.7912,
+      "gpu_window_seconds": null,
+      "inference_seconds": null
+    },
+    {
+      "model": "Whisper.cpp (large)",
+      "status": "error",
+      "client_wall_clock_seconds": 7.3045,
+      "gpu_window_seconds": null,
+      "inference_seconds": null
+    },
+    {
+      "model": "Whisper faster (large)",
+      "status": "ok",
+      "client_wall_clock_seconds": 32.4749,
+      "gpu_window_seconds": 26.2386,
+      "inference_seconds": 23.6107
+    },
+    {
+      "model": "NVIDIA Parakeet v3",
+      "status": "ok",
+      "client_wall_clock_seconds": 35.6289,
+      "gpu_window_seconds": 25.6806,
+      "inference_seconds": 1.6436
+    }
+  ]
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260221_183454/whisper.cpp_large.json

@@ -0,0 +1,6 @@
+{
+  "model": "Whisper.cpp (large)",
+  "status": "error",
+  "error": "Whisper.cpp requires model path. Set WHISPER_CPP_MODEL_LARGE or pass model_options_json={\"whisper_cpp_model\":\"/path/to/ggml-large-v3.bin\"}.",
+  "client_wall_clock_seconds": 7.3045
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260221_183454/whisper_large_v3.json

@@ -0,0 +1,6 @@
+{
+  "model": "Whisper Large V3",
+  "status": "error",
+  "error": "RuntimeError",
+  "client_wall_clock_seconds": 32.93
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260221_183454/whisper_large_v3_turbo.json

@@ -0,0 +1,6 @@
+{
+  "model": "Whisper Large V3 Turbo",
+  "status": "error",
+  "error": "TypeError",
+  "client_wall_clock_seconds": 49.7912
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260221_185214/benchmark_stats.json

@@ -0,0 +1,86 @@
+{
+  "space": "Ratnesh-dev/transcribe-diarize",
+  "audio_file": "Q3-FY26_5min.mp3",
+  "task": "transcribe",
+  "language": null,
+  "models": [
+    "Whisper Large V3",
+    "Whisper Large V3 Turbo",
+    "Whisper.cpp (large)",
+    "Whisper faster (large)",
+    "NVIDIA Parakeet v3"
+  ],
+  "benchmark_timing": {
+    "total_client_wall_clock_seconds": 172.2493
+  },
+  "leaderboard_by_gpu_window_seconds": [
+    {
+      "model": "NVIDIA Parakeet v3",
+      "api_name": "/transcribe_parakeet_v3",
+      "gpu_window_seconds": 25.1584,
+      "inference_seconds": 1.3924,
+      "client_wall_clock_seconds": 33.4613
+    },
+    {
+      "model": "Whisper faster (large)",
+      "api_name": "/transcribe_whisper_faster_large",
+      "gpu_window_seconds": 27.166,
+      "inference_seconds": 19.993,
+      "client_wall_clock_seconds": 32.5888
+    },
+    {
+      "model": "Whisper Large V3 Turbo",
+      "api_name": "/transcribe_whisper_large_v3_turbo",
+      "gpu_window_seconds": 27.5446,
+      "inference_seconds": 9.1112,
+      "client_wall_clock_seconds": 32.6999
+    }
+  ],
+  "summary": [
+    {
+      "model": "Whisper Large V3",
+      "status": "error",
+      "api_name": "/transcribe_whisper_large_v3",
+      "client_wall_clock_seconds": 67.0465,
+      "gpu_window_seconds": null,
+      "inference_seconds": null,
+      "error": "GPU task aborted"
+    },
+    {
+      "model": "Whisper Large V3 Turbo",
+      "status": "ok",
+      "api_name": "/transcribe_whisper_large_v3_turbo",
+      "client_wall_clock_seconds": 32.6999,
+      "gpu_window_seconds": 27.5446,
+      "inference_seconds": 9.1112,
+      "error": null
+    },
+    {
+      "model": "Whisper.cpp (large)",
+      "status": "error",
+      "api_name": "/transcribe_whisper_cpp_large",
+      "client_wall_clock_seconds": 6.4526,
+      "gpu_window_seconds": null,
+      "inference_seconds": null,
+      "error": "Whisper.cpp requires model path. Set WHISPER_CPP_MODEL_LARGE or pass model_options_json={\"whisper_cpp_model\":\"/path/to/ggml-large-v3.bin\"}."
+    },
+    {
+      "model": "Whisper faster (large)",
+      "status": "ok",
+      "api_name": "/transcribe_whisper_faster_large",
+      "client_wall_clock_seconds": 32.5888,
+      "gpu_window_seconds": 27.166,
+      "inference_seconds": 19.993,
+      "error": null
+    },
+    {
+      "model": "NVIDIA Parakeet v3",
+      "status": "ok",
+      "api_name": "/transcribe_parakeet_v3",
+      "client_wall_clock_seconds": 33.4613,
+      "gpu_window_seconds": 25.1584,
+      "inference_seconds": 1.3924,
+      "error": null
+    }
+  ]
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260221_185214/whisper.cpp_large.json

@@ -0,0 +1,7 @@
+{
+  "model": "Whisper.cpp (large)",
+  "api_name": "/transcribe_whisper_cpp_large",
+  "status": "error",
+  "client_wall_clock_seconds": 6.4526,
+  "error": "Whisper.cpp requires model path. Set WHISPER_CPP_MODEL_LARGE or pass model_options_json={\"whisper_cpp_model\":\"/path/to/ggml-large-v3.bin\"}."
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260221_185214/whisper_large_v3.json

@@ -0,0 +1,7 @@
+{
+  "model": "Whisper Large V3",
+  "api_name": "/transcribe_whisper_large_v3",
+  "status": "error",
+  "client_wall_clock_seconds": 67.0465,
+  "error": "GPU task aborted"
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_102228/benchmark_stats.json

@@ -0,0 +1,93 @@
+{
+  "space": "Ratnesh-dev/transcribe-diarize",
+  "audio_file": "Q3-FY26_5min.mp3",
+  "task": "transcribe",
+  "language": null,
+  "models": [
+    "Whisper Large V3",
+    "Whisper Large V3 Turbo",
+    "Whisper.cpp (large)",
+    "Whisper faster (large)",
+    "NVIDIA Parakeet v3"
+  ],
+  "benchmark_timing": {
+    "total_client_wall_clock_seconds": 195.695
+  },
+  "leaderboard_by_gpu_window_seconds": [
+    {
+      "model": "Whisper Large V3 Turbo",
+      "api_name": "/transcribe_whisper_large_v3_turbo",
+      "gpu_window_seconds": 21.8813,
+      "inference_seconds": 9.9142,
+      "client_wall_clock_seconds": 31.3042
+    },
+    {
+      "model": "Whisper faster (large)",
+      "api_name": "/transcribe_whisper_faster_large",
+      "gpu_window_seconds": 27.4783,
+      "inference_seconds": 13.5596,
+      "client_wall_clock_seconds": 34.0397
+    },
+    {
+      "model": "NVIDIA Parakeet v3",
+      "api_name": "/transcribe_parakeet_v3",
+      "gpu_window_seconds": 40.0005,
+      "inference_seconds": 2.0911,
+      "client_wall_clock_seconds": 49.2008
+    },
+    {
+      "model": "Whisper Large V3",
+      "api_name": "/transcribe_whisper_large_v3",
+      "gpu_window_seconds": 63.9729,
+      "inference_seconds": 42.212,
+      "client_wall_clock_seconds": 74.0941
+    }
+  ],
+  "summary": [
+    {
+      "model": "Whisper Large V3",
+      "status": "ok",
+      "api_name": "/transcribe_whisper_large_v3",
+      "client_wall_clock_seconds": 74.0941,
+      "gpu_window_seconds": 63.9729,
+      "inference_seconds": 42.212,
+      "error": null
+    },
+    {
+      "model": "Whisper Large V3 Turbo",
+      "status": "ok",
+      "api_name": "/transcribe_whisper_large_v3_turbo",
+      "client_wall_clock_seconds": 31.3042,
+      "gpu_window_seconds": 21.8813,
+      "inference_seconds": 9.9142,
+      "error": null
+    },
+    {
+      "model": "Whisper.cpp (large)",
+      "status": "error",
+      "api_name": "/transcribe_whisper_cpp_large",
+      "client_wall_clock_seconds": 7.0561,
+      "gpu_window_seconds": null,
+      "inference_seconds": null,
+      "error": "Whisper.cpp requires model path. Set WHISPER_CPP_MODEL_LARGE or pass model_options_json={\"whisper_cpp_model\":\"/path/to/ggml-large-v3.bin\"}."
+    },
+    {
+      "model": "Whisper faster (large)",
+      "status": "ok",
+      "api_name": "/transcribe_whisper_faster_large",
+      "client_wall_clock_seconds": 34.0397,
+      "gpu_window_seconds": 27.4783,
+      "inference_seconds": 13.5596,
+      "error": null
+    },
+    {
+      "model": "NVIDIA Parakeet v3",
+      "status": "ok",
+      "api_name": "/transcribe_parakeet_v3",
+      "client_wall_clock_seconds": 49.2008,
+      "gpu_window_seconds": 40.0005,
+      "inference_seconds": 2.0911,
+      "error": null
+    }
+  ]
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_102228/whisper.cpp_large.json

@@ -0,0 +1,7 @@
+{
+  "model": "Whisper.cpp (large)",
+  "api_name": "/transcribe_whisper_cpp_large",
+  "status": "error",
+  "client_wall_clock_seconds": 7.0561,
+  "error": "Whisper.cpp requires model path. Set WHISPER_CPP_MODEL_LARGE or pass model_options_json={\"whisper_cpp_model\":\"/path/to/ggml-large-v3.bin\"}."
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_155458/benchmark_stats.json

@@ -0,0 +1,42 @@
+{
+  "space": "Ratnesh-dev/transcribe-diarize",
+  "audio_file": "Q3-FY26_5min.mp3",
+  "task": "transcribe",
+  "language": null,
+  "models": [
+    "Whisper Large V3 Turbo",
+    "NVIDIA Parakeet v3"
+  ],
+  "benchmark_timing": {
+    "total_client_wall_clock_seconds": 51.6486
+  },
+  "leaderboard_by_gpu_window_seconds": [
+    {
+      "model": "NVIDIA Parakeet v3",
+      "api_name": "/transcribe_parakeet_v3",
+      "gpu_window_seconds": 27.4834,
+      "inference_seconds": 1.9277,
+      "client_wall_clock_seconds": 35.6015
+    }
+  ],
+  "summary": [
+    {
+      "model": "Whisper Large V3 Turbo",
+      "status": "error",
+      "api_name": "/transcribe_whisper_large_v3_turbo",
+      "client_wall_clock_seconds": 16.047,
+      "gpu_window_seconds": null,
+      "inference_seconds": null,
+      "error": "ValueError"
+    },
+    {
+      "model": "NVIDIA Parakeet v3",
+      "status": "ok",
+      "api_name": "/transcribe_parakeet_v3",
+      "client_wall_clock_seconds": 35.6015,
+      "gpu_window_seconds": 27.4834,
+      "inference_seconds": 1.9277,
+      "error": null
+    }
+  ]
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_155458/whisper_large_v3_turbo.json

@@ -0,0 +1,8 @@
+{
+  "model": "Whisper Large V3 Turbo",
+  "api_name": "/transcribe_whisper_large_v3_turbo",
+  "status": "error",
+  "client_wall_clock_seconds": 16.047,
+  "effective_model_options_json": "{\"chunk_length_s\": 30, \"batch_size\": 16, \"long_audio_threshold_seconds\": 120, \"num_beams\": 1}",
+  "error": "ValueError"
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_160801/benchmark_stats.json

@@ -0,0 +1,34 @@
+{
+  "space": "Ratnesh-dev/transcribe-diarize",
+  "audio_file": "Q3-FY26_5min.mp3",
+  "task": "transcribe",
+  "language": null,
+  "models": [
+    "Whisper Large V3 Turbo",
+    "NVIDIA Parakeet v3"
+  ],
+  "benchmark_timing": {
+    "total_client_wall_clock_seconds": 70.9442
+  },
+  "leaderboard_by_gpu_window_seconds": [],
+  "summary": [
+    {
+      "model": "Whisper Large V3 Turbo",
+      "status": "error",
+      "api_name": "/transcribe_whisper_large_v3_turbo",
+      "client_wall_clock_seconds": 64.672,
+      "gpu_window_seconds": null,
+      "inference_seconds": null,
+      "error": "IndexError"
+    },
+    {
+      "model": "NVIDIA Parakeet v3",
+      "status": "error",
+      "api_name": "/transcribe_parakeet_v3",
+      "client_wall_clock_seconds": 6.2719,
+      "gpu_window_seconds": null,
+      "inference_seconds": null,
+      "error": "You have exceeded your GPU quota (60s requested vs. 22s left). Try again in 23:46:12"
+    }
+  ]
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_160801/nvidia_parakeet_v3.json

@@ -0,0 +1,8 @@
+{
+  "model": "NVIDIA Parakeet v3",
+  "api_name": "/transcribe_parakeet_v3",
+  "status": "error",
+  "client_wall_clock_seconds": 6.2719,
+  "effective_model_options_json": "{\"batch_size\": 1, \"long_audio_threshold_seconds\": 480, \"local_attention_left\": 256, \"local_attention_right\": 256, \"subsampling_conv_chunking_factor\": 1}",
+  "error": "You have exceeded your GPU quota (60s requested vs. 22s left). Try again in 23:46:12"
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_160801/whisper_large_v3_turbo.json

@@ -0,0 +1,8 @@
+{
+  "model": "Whisper Large V3 Turbo",
+  "api_name": "/transcribe_whisper_large_v3_turbo",
+  "status": "error",
+  "client_wall_clock_seconds": 64.672,
+  "effective_model_options_json": "{\"chunk_length_s\": 30, \"batch_size\": 16, \"long_audio_threshold_seconds\": 120, \"num_beams\": 1}",
+  "error": "IndexError"
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_162404/benchmark_stats.json

@@ -0,0 +1,34 @@
+{
+  "space": "Ratnesh-dev/transcribe-diarize",
+  "audio_file": "Q3-FY26_5min.mp3",
+  "task": "transcribe",
+  "language": null,
+  "models": [
+    "Whisper Large V3 Turbo",
+    "NVIDIA Parakeet v3"
+  ],
+  "benchmark_timing": {
+    "total_client_wall_clock_seconds": 12.4255
+  },
+  "leaderboard_by_gpu_window_seconds": [],
+  "summary": [
+    {
+      "model": "Whisper Large V3 Turbo",
+      "status": "error",
+      "api_name": "/transcribe_whisper_large_v3_turbo",
+      "client_wall_clock_seconds": 5.9353,
+      "gpu_window_seconds": null,
+      "inference_seconds": null,
+      "error": "You have exceeded your GPU quota (60s requested vs. 22s left). Try again in 23:30:15"
+    },
+    {
+      "model": "NVIDIA Parakeet v3",
+      "status": "error",
+      "api_name": "/transcribe_parakeet_v3",
+      "client_wall_clock_seconds": 6.4901,
+      "gpu_window_seconds": null,
+      "inference_seconds": null,
+      "error": "You have exceeded your GPU quota (60s requested vs. 22s left). Try again in 23:30:10"
+    }
+  ]
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_162404/nvidia_parakeet_v3.json

@@ -0,0 +1,8 @@
+{
+  "model": "NVIDIA Parakeet v3",
+  "api_name": "/transcribe_parakeet_v3",
+  "status": "error",
+  "client_wall_clock_seconds": 6.4901,
+  "effective_model_options_json": "{\"batch_size\": 1, \"long_audio_threshold_seconds\": 480, \"local_attention_left\": 256, \"local_attention_right\": 256, \"subsampling_conv_chunking_factor\": 1}",
+  "error": "You have exceeded your GPU quota (60s requested vs. 22s left). Try again in 23:30:10"
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_162404/whisper_large_v3_turbo.json

@@ -0,0 +1,8 @@
+{
+  "model": "Whisper Large V3 Turbo",
+  "api_name": "/transcribe_whisper_large_v3_turbo",
+  "status": "error",
+  "client_wall_clock_seconds": 5.9353,
+  "effective_model_options_json": "{\"chunk_length_s\": 30, \"batch_size\": 16, \"long_audio_threshold_seconds\": 120, \"num_beams\": 1}",
+  "error": "You have exceeded your GPU quota (60s requested vs. 22s left). Try again in 23:30:15"
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_162651/benchmark_stats.json

@@ -0,0 +1,42 @@
+{
+  "space": "Ratnesh-dev/transcribe-diarize",
+  "audio_file": "Q3-FY26_5min.mp3",
+  "task": "transcribe",
+  "language": null,
+  "models": [
+    "Whisper Large V3 Turbo",
+    "NVIDIA Parakeet v3"
+  ],
+  "benchmark_timing": {
+    "total_client_wall_clock_seconds": 105.8342
+  },
+  "leaderboard_by_gpu_window_seconds": [
+    {
+      "model": "NVIDIA Parakeet v3",
+      "api_name": "/transcribe_parakeet_v3",
+      "gpu_window_seconds": 28.6758,
+      "inference_seconds": 2.5322,
+      "client_wall_clock_seconds": 36.9385
+    }
+  ],
+  "summary": [
+    {
+      "model": "Whisper Large V3 Turbo",
+      "status": "error",
+      "api_name": "/transcribe_whisper_large_v3_turbo",
+      "client_wall_clock_seconds": 68.8956,
+      "gpu_window_seconds": null,
+      "inference_seconds": null,
+      "error": "GPU task aborted"
+    },
+    {
+      "model": "NVIDIA Parakeet v3",
+      "status": "ok",
+      "api_name": "/transcribe_parakeet_v3",
+      "client_wall_clock_seconds": 36.9385,
+      "gpu_window_seconds": 28.6758,
+      "inference_seconds": 2.5322,
+      "error": null
+    }
+  ]
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_162651/whisper_large_v3_turbo.json

@@ -0,0 +1,8 @@
+{
+  "model": "Whisper Large V3 Turbo",
+  "api_name": "/transcribe_whisper_large_v3_turbo",
+  "status": "error",
+  "client_wall_clock_seconds": 68.8956,
+  "effective_model_options_json": "{\"chunk_length_s\": 30, \"batch_size\": 16, \"long_audio_threshold_seconds\": 120, \"num_beams\": 1}",
+  "error": "GPU task aborted"
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_163912/benchmark_stats.json

@@ -0,0 +1,42 @@
+{
+  "space": "Ratnesh-dev/transcribe-diarize",
+  "audio_file": "Q3-FY26_5min.mp3",
+  "task": "transcribe",
+  "language": null,
+  "models": [
+    "Whisper Large V3 Turbo",
+    "NVIDIA Parakeet v3"
+  ],
+  "benchmark_timing": {
+    "total_client_wall_clock_seconds": 50.2386
+  },
+  "leaderboard_by_gpu_window_seconds": [
+    {
+      "model": "NVIDIA Parakeet v3",
+      "api_name": "/transcribe_parakeet_v3",
+      "gpu_window_seconds": 26.4022,
+      "inference_seconds": 1.9466,
+      "client_wall_clock_seconds": 35.5696
+    }
+  ],
+  "summary": [
+    {
+      "model": "Whisper Large V3 Turbo",
+      "status": "error",
+      "api_name": "/transcribe_whisper_large_v3_turbo",
+      "client_wall_clock_seconds": 14.6688,
+      "gpu_window_seconds": null,
+      "inference_seconds": null,
+      "error": "TypeError"
+    },
+    {
+      "model": "NVIDIA Parakeet v3",
+      "status": "ok",
+      "api_name": "/transcribe_parakeet_v3",
+      "client_wall_clock_seconds": 35.5696,
+      "gpu_window_seconds": 26.4022,
+      "inference_seconds": 1.9466,
+      "error": null
+    }
+  ]
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_163912/whisper_large_v3_turbo.json

@@ -0,0 +1,8 @@
+{
+  "model": "Whisper Large V3 Turbo",
+  "api_name": "/transcribe_whisper_large_v3_turbo",
+  "status": "error",
+  "client_wall_clock_seconds": 14.6688,
+  "effective_model_options_json": "{\"chunk_length_s\": 30, \"batch_size\": 16, \"long_audio_threshold_seconds\": 120, \"num_beams\": 1}",
+  "error": "TypeError"
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_170425/benchmark_stats.json

@@ -0,0 +1,42 @@
+{
+  "space": "Ratnesh-dev/transcribe-diarize",
+  "audio_file": "Q3-FY26_5min.mp3",
+  "task": "transcribe",
+  "language": null,
+  "models": [
+    "Whisper Large V3 Turbo",
+    "NVIDIA Parakeet v3"
+  ],
+  "benchmark_timing": {
+    "total_client_wall_clock_seconds": 19.1928
+  },
+  "leaderboard_by_gpu_window_seconds": [
+    {
+      "model": "NVIDIA Parakeet v3",
+      "api_name": "/transcribe_parakeet_v3",
+      "gpu_window_seconds": 1.7501,
+      "inference_seconds": 1.7462,
+      "client_wall_clock_seconds": 10.4831
+    }
+  ],
+  "summary": [
+    {
+      "model": "Whisper Large V3 Turbo",
+      "status": "error",
+      "api_name": "/transcribe_whisper_large_v3_turbo",
+      "client_wall_clock_seconds": 8.7096,
+      "gpu_window_seconds": null,
+      "inference_seconds": null,
+      "error": "TypeError"
+    },
+    {
+      "model": "NVIDIA Parakeet v3",
+      "status": "ok",
+      "api_name": "/transcribe_parakeet_v3",
+      "client_wall_clock_seconds": 10.4831,
+      "gpu_window_seconds": 1.7501,
+      "inference_seconds": 1.7462,
+      "error": null
+    }
+  ]
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_170425/whisper_large_v3_turbo.json

@@ -0,0 +1,8 @@
+{
+  "model": "Whisper Large V3 Turbo",
+  "api_name": "/transcribe_whisper_large_v3_turbo",
+  "status": "error",
+  "client_wall_clock_seconds": 8.7096,
+  "effective_model_options_json": "{\"num_beams\": 1, \"temperature\": 0.0}",
+  "error": "TypeError"
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_171344/benchmark_stats.json

@@ -0,0 +1,49 @@
+{
+  "space": "Ratnesh-dev/transcribe-diarize",
+  "audio_file": "Q3-FY26_5min.mp3",
+  "task": "transcribe",
+  "language": null,
+  "models": [
+    "Whisper Large V3 Turbo",
+    "NVIDIA Parakeet v3"
+  ],
+  "benchmark_timing": {
+    "total_client_wall_clock_seconds": 57.7283
+  },
+  "leaderboard_by_gpu_window_seconds": [
+    {
+      "model": "NVIDIA Parakeet v3",
+      "api_name": "/transcribe_parakeet_v3",
+      "gpu_window_seconds": 2.2344,
+      "inference_seconds": 2.221,
+      "client_wall_clock_seconds": 15.0925
+    },
+    {
+      "model": "Whisper Large V3 Turbo",
+      "api_name": "/transcribe_whisper_large_v3_turbo",
+      "gpu_window_seconds": 35.4609,
+      "inference_seconds": 35.4596,
+      "client_wall_clock_seconds": 42.6357
+    }
+  ],
+  "summary": [
+    {
+      "model": "Whisper Large V3 Turbo",
+      "status": "ok",
+      "api_name": "/transcribe_whisper_large_v3_turbo",
+      "client_wall_clock_seconds": 42.6357,
+      "gpu_window_seconds": 35.4609,
+      "inference_seconds": 35.4596,
+      "error": null
+    },
+    {
+      "model": "NVIDIA Parakeet v3",
+      "status": "ok",
+      "api_name": "/transcribe_parakeet_v3",
+      "client_wall_clock_seconds": 15.0925,
+      "gpu_window_seconds": 2.2344,
+      "inference_seconds": 2.221,
+      "error": null
+    }
+  ]
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_171344/whisper_large_v3_turbo.json

@@ -0,0 +1,4815 @@
+{
+  "model": "Whisper Large V3 Turbo",
+  "task": "transcribe",
+  "audio_file": "/tmp/gradio/7805a61018caf7c7cf375cda35bf3e0e8c88096aaec4ccbf416d6be775af26aa/Q3-FY26_5min.mp3",
+  "postprocess_prompt": "You have the transcript of the quarterly conference call for Kotak Mahindra BankClean up any mispelled words, false starts, repeated words from the transcript.Do not add anything new on your own. Here are some executive names which might be misspelled:  Ashok Vaswani Jaideep Hansraj Paritosh Kashyap Anup Kumar Saha Devang Gheewalla Manish Agarwal Manish Kothari Oisharya Das Pranav Mishra Vyomesh Kapasi",
+  "model_options": {
+    "num_beams": 1,
+    "temperature": 0.0
+  },
+  "zerogpu_timing": {
+    "gpu_window_seconds": 35.4609,
+    "inference_seconds": 35.4596
+  },
+  "raw_output": {
+    "backend": "transformers-pipeline-basic",
+    "model_name": "openai/whisper-large-v3-turbo",
+    "prompt": {
+      "initial_prompt_provided": true
+    },
+    "result": {
+      "text": " Ladies and gentlemen, good day and welcome to the Kortak Mahindra Bank Q3 FY26 Earnings Conference Call. As a reminder, all participant lines will be in the listen-only mode and there will be an opportunity for you to ask questions after the presentation concludes. Should you need assistance during this conference call, please signal an operator by pressing star and then zero on your touchstone phone. Please note that this conference is being recorded. I now hand the conference over to Mr. Ashok Vaswani, Managing Director and CEO of Kotak Mahindra Bank. Thank you and over to you, sir. Thank you. Thank you so much, operator. Good evening. Good evening, everyone, and thank you for joining us. Before I begin, let me welcome Anup Saha, our new WTD designate. Anup brings over three decades of vast experience and domain expertise across many fields, including retail finance, analytics, risk, collections, and financial service operations. He will oversee our consumer bank, data analytics, and marketing verticals. His additions strengthens our leadership team and accelerates our ability to execute at scale. Moving on to today's agenda, I'll make some opening remarks. Devang will cover the numbers in detail. Pradab, the deposit franchise, Yomesh, consumer assets, Manish, commercial banking, Paritosh, wholesale banking and institutional businesses, and finally, Jerry will update you on our subs. Let me first talk a little bit about the macroeconomic environment. The global landscape has become more volatile. Geopolitical tensions, trade and tariff uncertainties have weighed on global investor confidence. While there seems to be a high level of confidence to get a trade deal with the EU, the absence of a final trade agreement with the US continues to add to this uncertainty. We have seen unprecedented strengthening of gold and silver and outflows from India by FIIs to other markets that have perceived to be more attractive to foreign investors. However, the Indian economy continues to demonstrate strong and stable growth. GDP growth continues to be resilient with a positive outlook. This is supported by several enabling actions like the RBI's repo rate cuts and GST rate rationalization. 3Q also saw an uptick in growth led by the usual festive demand. Rural India overall looks stable, supported by good monsuits. The rupee weakness has partly offset tariff impacts, but has also affected foreign flows and liquidity. With respect to capital flows, while FIIs have been net sellers, domestic institutional investors have provided support with around $24 billion in inflows in Q3 FY26, reinforcing the savers to investors' strength. This is also evident in the strong IPO pipeline expected for calendar 2026. In the banking sector specifically, system credit growth has been healthy, but the flows into commodities and capital markets are putting more pressure on low-cost deposits. There is enhanced volatility seen in the banking sector liquidity. While benchmark rates are reduced, longer-term credit yields have gone down. Within this context, coming to quota, we remain focused on our strategy to scale responsibly. This discipline is reflected in our quarterly performance. In the bank, net advances grew 16% YOY. This is in line with our stated philosophy of growing our advances in the range of 1.5 to 2 times nominal GDP growth. We have seen consistent growth of around 4% per quarter for the last three quarters of this financial year. Having deposits grew 15% year on year, we have been focusing on granular casa growth. This is on the back of our focus segment strategy, and I will come to that shortly. We called out unsecured retail loans earlier. We said we would grow gradually and responsibly. with the unsecured assets growing in absolute terms in the first instance. And that's exactly what we have done and will continue to do so. The bank's NIL continues to be healthy at 4.54%, supported by a low cost of funds. Credit costs, as indicated previously, continue the downward trend, supported by updated underwriting models and enhanced collection efforts. Credit costs are down sequentially from 93 basis points in Q1 to 79 basis points in Q1.",
+      "chunks": [
+        {
+          "text": " Ladies",
+          "timestamp": [
+            0.52,
+            0.8
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            0.8,
+            1.0
+          ]
+        },
+        {
+          "text": " gentlemen,",
+          "timestamp": [
+            1.0,
+            1.72
+          ]
+        },
+        {
+          "text": " good",
+          "timestamp": [
+            1.72,
+            2.08
+          ]
+        },
+        {
+          "text": " day",
+          "timestamp": [
+            2.08,
+            2.3
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            2.3,
+            2.5
+          ]
+        },
+        {
+          "text": " welcome",
+          "timestamp": [
+            2.5,
+            2.88
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            2.88,
+            3.28
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            3.28,
+            3.38
+          ]
+        },
+        {
+          "text": " Kortak",
+          "timestamp": [
+            3.38,
+            3.72
+          ]
+        },
+        {
+          "text": " Mahindra",
+          "timestamp": [
+            3.72,
+            4.12
+          ]
+        },
+        {
+          "text": " Bank",
+          "timestamp": [
+            4.12,
+            4.44
+          ]
+        },
+        {
+          "text": " Q3",
+          "timestamp": [
+            4.44,
+            5.1
+          ]
+        },
+        {
+          "text": " FY26",
+          "timestamp": [
+            5.1,
+            6.0
+          ]
+        },
+        {
+          "text": " Earnings",
+          "timestamp": [
+            6.0,
+            6.52
+          ]
+        },
+        {
+          "text": " Conference",
+          "timestamp": [
+            6.52,
+            6.84
+          ]
+        },
+        {
+          "text": " Call.",
+          "timestamp": [
+            6.84,
+            7.32
+          ]
+        },
+        {
+          "text": " As",
+          "timestamp": [
+            8.84,
+            9.14
+          ]
+        },
+        {
+          "text": " a",
+          "timestamp": [
+            9.14,
+            9.22
+          ]
+        },
+        {
+          "text": " reminder,",
+          "timestamp": [
+            9.22,
+            9.88
+          ]
+        },
+        {
+          "text": " all",
+          "timestamp": [
+            9.88,
+            10.36
+          ]
+        },
+        {
+          "text": " participant",
+          "timestamp": [
+            10.36,
+            10.84
+          ]
+        },
+        {
+          "text": " lines",
+          "timestamp": [
+            10.84,
+            11.18
+          ]
+        },
+        {
+          "text": " will",
+          "timestamp": [
+            11.18,
+            11.46
+          ]
+        },
+        {
+          "text": " be",
+          "timestamp": [
+            11.46,
+            11.6
+          ]
+        },
+        {
+          "text": " in",
+          "timestamp": [
+            11.6,
+            11.7
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            11.7,
+            11.8
+          ]
+        },
+        {
+          "text": " listen",
+          "timestamp": [
+            11.8,
+            12.02
+          ]
+        },
+        {
+          "text": "-only",
+          "timestamp": [
+            12.02,
+            12.28
+          ]
+        },
+        {
+          "text": " mode",
+          "timestamp": [
+            12.28,
+            12.62
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            12.62,
+            13.2
+          ]
+        },
+        {
+          "text": " there",
+          "timestamp": [
+            13.2,
+            13.32
+          ]
+        },
+        {
+          "text": " will",
+          "timestamp": [
+            13.32,
+            13.44
+          ]
+        },
+        {
+          "text": " be",
+          "timestamp": [
+            13.44,
+            13.54
+          ]
+        },
+        {
+          "text": " an",
+          "timestamp": [
+            13.54,
+            13.68
+          ]
+        },
+        {
+          "text": " opportunity",
+          "timestamp": [
+            13.68,
+            14.02
+          ]
+        },
+        {
+          "text": " for",
+          "timestamp": [
+            14.02,
+            14.38
+          ]
+        },
+        {
+          "text": " you",
+          "timestamp": [
+            14.38,
+            14.48
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            14.48,
+            14.6
+          ]
+        },
+        {
+          "text": " ask",
+          "timestamp": [
+            14.6,
+            14.9
+          ]
+        },
+        {
+          "text": " questions",
+          "timestamp": [
+            14.9,
+            15.46
+          ]
+        },
+        {
+          "text": " after",
+          "timestamp": [
+            15.46,
+            16.26
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            16.26,
+            16.4
+          ]
+        },
+        {
+          "text": " presentation",
+          "timestamp": [
+            16.4,
+            16.9
+          ]
+        },
+        {
+          "text": " concludes.",
+          "timestamp": [
+            16.9,
+            17.92
+          ]
+        },
+        {
+          "text": " Should",
+          "timestamp": [
+            18.6,
+            18.8
+          ]
+        },
+        {
+          "text": " you",
+          "timestamp": [
+            18.8,
+            18.92
+          ]
+        },
+        {
+          "text": " need",
+          "timestamp": [
+            18.92,
+            19.08
+          ]
+        },
+        {
+          "text": " assistance",
+          "timestamp": [
+            19.08,
+            19.52
+          ]
+        },
+        {
+          "text": " during",
+          "timestamp": [
+            19.52,
+            19.8
+          ]
+        },
+        {
+          "text": " this",
+          "timestamp": [
+            19.8,
+            20.04
+          ]
+        },
+        {
+          "text": " conference",
+          "timestamp": [
+            20.04,
+            20.38
+          ]
+        },
+        {
+          "text": " call,",
+          "timestamp": [
+            20.38,
+            20.98
+          ]
+        },
+        {
+          "text": " please",
+          "timestamp": [
+            20.98,
+            21.36
+          ]
+        },
+        {
+          "text": " signal",
+          "timestamp": [
+            21.36,
+            21.64
+          ]
+        },
+        {
+          "text": " an",
+          "timestamp": [
+            21.64,
+            21.82
+          ]
+        },
+        {
+          "text": " operator",
+          "timestamp": [
+            21.82,
+            22.2
+          ]
+        },
+        {
+          "text": " by",
+          "timestamp": [
+            22.2,
+            22.62
+          ]
+        },
+        {
+          "text": " pressing",
+          "timestamp": [
+            22.62,
+            22.94
+          ]
+        },
+        {
+          "text": " star",
+          "timestamp": [
+            22.94,
+            23.28
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            23.28,
+            23.62
+          ]
+        },
+        {
+          "text": " then",
+          "timestamp": [
+            23.62,
+            23.8
+          ]
+        },
+        {
+          "text": " zero",
+          "timestamp": [
+            23.8,
+            24.18
+          ]
+        },
+        {
+          "text": " on",
+          "timestamp": [
+            24.18,
+            24.74
+          ]
+        },
+        {
+          "text": " your",
+          "timestamp": [
+            24.74,
+            24.88
+          ]
+        },
+        {
+          "text": " touchstone",
+          "timestamp": [
+            24.88,
+            25.32
+          ]
+        },
+        {
+          "text": " phone.",
+          "timestamp": [
+            25.32,
+            25.96
+          ]
+        },
+        {
+          "text": " Please",
+          "timestamp": [
+            26.64,
+            27.02
+          ]
+        },
+        {
+          "text": " note",
+          "timestamp": [
+            27.02,
+            27.34
+          ]
+        },
+        {
+          "text": " that",
+          "timestamp": [
+            27.34,
+            27.84
+          ]
+        },
+        {
+          "text": " this",
+          "timestamp": [
+            27.84,
+            28.0
+          ]
+        },
+        {
+          "text": " conference",
+          "timestamp": [
+            28.0,
+            28.36
+          ]
+        },
+        {
+          "text": " is",
+          "timestamp": [
+            28.36,
+            28.66
+          ]
+        },
+        {
+          "text": " being",
+          "timestamp": [
+            28.66,
+            28.9
+          ]
+        },
+        {
+          "text": " recorded.",
+          "timestamp": [
+            28.9,
+            29.56
+          ]
+        },
+        {
+          "text": " I",
+          "timestamp": [
+            30.5,
+            30.76
+          ]
+        },
+        {
+          "text": " now",
+          "timestamp": [
+            30.76,
+            30.98
+          ]
+        },
+        {
+          "text": " hand",
+          "timestamp": [
+            30.98,
+            31.2
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            31.2,
+            31.32
+          ]
+        },
+        {
+          "text": " conference",
+          "timestamp": [
+            31.32,
+            31.7
+          ]
+        },
+        {
+          "text": " over",
+          "timestamp": [
+            31.7,
+            32.04
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            32.04,
+            32.44
+          ]
+        },
+        {
+          "text": " Mr.",
+          "timestamp": [
+            32.44,
+            32.74
+          ]
+        },
+        {
+          "text": " Ashok",
+          "timestamp": [
+            32.74,
+            32.98
+          ]
+        },
+        {
+          "text": " Vaswani,",
+          "timestamp": [
+            32.98,
+            33.92
+          ]
+        },
+        {
+          "text": " Managing",
+          "timestamp": [
+            33.92,
+            34.36
+          ]
+        },
+        {
+          "text": " Director",
+          "timestamp": [
+            34.36,
+            34.76
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            34.76,
+            35.2
+          ]
+        },
+        {
+          "text": " CEO",
+          "timestamp": [
+            35.2,
+            35.66
+          ]
+        },
+        {
+          "text": " of",
+          "timestamp": [
+            35.66,
+            36.24
+          ]
+        },
+        {
+          "text": " Kotak",
+          "timestamp": [
+            36.24,
+            36.66
+          ]
+        },
+        {
+          "text": " Mahindra",
+          "timestamp": [
+            36.66,
+            37.04
+          ]
+        },
+        {
+          "text": " Bank.",
+          "timestamp": [
+            37.04,
+            37.52
+          ]
+        },
+        {
+          "text": " Thank",
+          "timestamp": [
+            38.04,
+            38.18
+          ]
+        },
+        {
+          "text": " you",
+          "timestamp": [
+            38.18,
+            38.38
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            38.38,
+            38.72
+          ]
+        },
+        {
+          "text": " over",
+          "timestamp": [
+            38.72,
+            39.02
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            39.02,
+            39.14
+          ]
+        },
+        {
+          "text": " you,",
+          "timestamp": [
+            39.14,
+            39.38
+          ]
+        },
+        {
+          "text": " sir.",
+          "timestamp": [
+            39.38,
+            39.78
+          ]
+        },
+        {
+          "text": " Thank",
+          "timestamp": [
+            40.8,
+            40.92
+          ]
+        },
+        {
+          "text": " you.",
+          "timestamp": [
+            40.92,
+            41.2
+          ]
+        },
+        {
+          "text": " Thank",
+          "timestamp": [
+            41.2,
+            41.32
+          ]
+        },
+        {
+          "text": " you",
+          "timestamp": [
+            41.32,
+            41.44
+          ]
+        },
+        {
+          "text": " so",
+          "timestamp": [
+            41.44,
+            41.6
+          ]
+        },
+        {
+          "text": " much,",
+          "timestamp": [
+            41.6,
+            41.86
+          ]
+        },
+        {
+          "text": " operator.",
+          "timestamp": [
+            41.86,
+            42.46
+          ]
+        },
+        {
+          "text": " Good",
+          "timestamp": [
+            42.9,
+            43.48
+          ]
+        },
+        {
+          "text": " evening.",
+          "timestamp": [
+            43.48,
+            43.96
+          ]
+        },
+        {
+          "text": " Good",
+          "timestamp": [
+            43.96,
+            44.1
+          ]
+        },
+        {
+          "text": " evening,",
+          "timestamp": [
+            44.1,
+            44.4
+          ]
+        },
+        {
+          "text": " everyone,",
+          "timestamp": [
+            44.4,
+            44.72
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            44.72,
+            44.9
+          ]
+        },
+        {
+          "text": " thank",
+          "timestamp": [
+            44.9,
+            45.12
+          ]
+        },
+        {
+          "text": " you",
+          "timestamp": [
+            45.12,
+            45.24
+          ]
+        },
+        {
+          "text": " for",
+          "timestamp": [
+            45.24,
+            45.4
+          ]
+        },
+        {
+          "text": " joining",
+          "timestamp": [
+            45.4,
+            45.68
+          ]
+        },
+        {
+          "text": " us.",
+          "timestamp": [
+            45.68,
+            46.18
+          ]
+        },
+        {
+          "text": " Before",
+          "timestamp": [
+            46.72,
+            47.04
+          ]
+        },
+        {
+          "text": " I",
+          "timestamp": [
+            47.04,
+            47.2
+          ]
+        },
+        {
+          "text": " begin,",
+          "timestamp": [
+            47.2,
+            47.8
+          ]
+        },
+        {
+          "text": " let",
+          "timestamp": [
+            47.8,
+            48.22
+          ]
+        },
+        {
+          "text": " me",
+          "timestamp": [
+            48.22,
+            48.4
+          ]
+        },
+        {
+          "text": " welcome",
+          "timestamp": [
+            48.4,
+            48.88
+          ]
+        },
+        {
+          "text": " Anup",
+          "timestamp": [
+            48.88,
+            49.3
+          ]
+        },
+        {
+          "text": " Saha,",
+          "timestamp": [
+            49.3,
+            50.06
+          ]
+        },
+        {
+          "text": " our",
+          "timestamp": [
+            50.06,
+            50.26
+          ]
+        },
+        {
+          "text": " new",
+          "timestamp": [
+            50.26,
+            50.54
+          ]
+        },
+        {
+          "text": " WTD",
+          "timestamp": [
+            50.54,
+            52.48
+          ]
+        },
+        {
+          "text": " designate.",
+          "timestamp": [
+            52.48,
+            53.22
+          ]
+        },
+        {
+          "text": " Anup",
+          "timestamp": [
+            53.6,
+            53.9
+          ]
+        },
+        {
+          "text": " brings",
+          "timestamp": [
+            53.9,
+            54.14
+          ]
+        },
+        {
+          "text": " over",
+          "timestamp": [
+            54.14,
+            54.38
+          ]
+        },
+        {
+          "text": " three",
+          "timestamp": [
+            54.38,
+            54.6
+          ]
+        },
+        {
+          "text": " decades",
+          "timestamp": [
+            54.6,
+            54.96
+          ]
+        },
+        {
+          "text": " of",
+          "timestamp": [
+            54.96,
+            55.16
+          ]
+        },
+        {
+          "text": " vast",
+          "timestamp": [
+            55.16,
+            55.46
+          ]
+        },
+        {
+          "text": " experience",
+          "timestamp": [
+            55.46,
+            55.96
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            55.96,
+            56.62
+          ]
+        },
+        {
+          "text": " domain",
+          "timestamp": [
+            56.62,
+            56.94
+          ]
+        },
+        {
+          "text": " expertise",
+          "timestamp": [
+            56.94,
+            57.46
+          ]
+        },
+        {
+          "text": " across",
+          "timestamp": [
+            57.46,
+            58.08
+          ]
+        },
+        {
+          "text": " many",
+          "timestamp": [
+            58.08,
+            58.48
+          ]
+        },
+        {
+          "text": " fields,",
+          "timestamp": [
+            58.48,
+            59.04
+          ]
+        },
+        {
+          "text": " including",
+          "timestamp": [
+            59.38,
+            59.78
+          ]
+        },
+        {
+          "text": " retail",
+          "timestamp": [
+            59.78,
+            60.12
+          ]
+        },
+        {
+          "text": " finance,",
+          "timestamp": [
+            60.12,
+            60.84
+          ]
+        },
+        {
+          "text": " analytics,",
+          "timestamp": [
+            60.84,
+            61.64
+          ]
+        },
+        {
+          "text": " risk,",
+          "timestamp": [
+            61.64,
+            62.62
+          ]
+        },
+        {
+          "text": " collections,",
+          "timestamp": [
+            62.62,
+            63.3
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            63.3,
+            63.54
+          ]
+        },
+        {
+          "text": " financial",
+          "timestamp": [
+            63.54,
+            63.88
+          ]
+        },
+        {
+          "text": " service",
+          "timestamp": [
+            63.88,
+            64.24
+          ]
+        },
+        {
+          "text": " operations.",
+          "timestamp": [
+            64.24,
+            65.04
+          ]
+        },
+        {
+          "text": " He",
+          "timestamp": [
+            65.58,
+            65.74
+          ]
+        },
+        {
+          "text": " will",
+          "timestamp": [
+            65.74,
+            65.88
+          ]
+        },
+        {
+          "text": " oversee",
+          "timestamp": [
+            65.88,
+            66.28
+          ]
+        },
+        {
+          "text": " our",
+          "timestamp": [
+            66.28,
+            66.46
+          ]
+        },
+        {
+          "text": " consumer",
+          "timestamp": [
+            66.46,
+            66.82
+          ]
+        },
+        {
+          "text": " bank,",
+          "timestamp": [
+            66.82,
+            67.7
+          ]
+        },
+        {
+          "text": " data",
+          "timestamp": [
+            67.7,
+            67.9
+          ]
+        },
+        {
+          "text": " analytics,",
+          "timestamp": [
+            67.9,
+            68.58
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            68.58,
+            68.8
+          ]
+        },
+        {
+          "text": " marketing",
+          "timestamp": [
+            68.8,
+            69.22
+          ]
+        },
+        {
+          "text": " verticals.",
+          "timestamp": [
+            69.22,
+            70.04
+          ]
+        },
+        {
+          "text": " His",
+          "timestamp": [
+            70.4,
+            70.6
+          ]
+        },
+        {
+          "text": " additions",
+          "timestamp": [
+            70.6,
+            71.02
+          ]
+        },
+        {
+          "text": " strengthens",
+          "timestamp": [
+            71.02,
+            71.64
+          ]
+        },
+        {
+          "text": " our",
+          "timestamp": [
+            71.64,
+            71.78
+          ]
+        },
+        {
+          "text": " leadership",
+          "timestamp": [
+            71.78,
+            72.1
+          ]
+        },
+        {
+          "text": " team",
+          "timestamp": [
+            72.1,
+            72.48
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            72.48,
+            73.1
+          ]
+        },
+        {
+          "text": " accelerates",
+          "timestamp": [
+            73.1,
+            73.78
+          ]
+        },
+        {
+          "text": " our",
+          "timestamp": [
+            73.78,
+            73.92
+          ]
+        },
+        {
+          "text": " ability",
+          "timestamp": [
+            73.92,
+            74.32
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            74.32,
+            74.96
+          ]
+        },
+        {
+          "text": " execute",
+          "timestamp": [
+            74.96,
+            75.56
+          ]
+        },
+        {
+          "text": " at",
+          "timestamp": [
+            75.56,
+            76.0
+          ]
+        },
+        {
+          "text": " scale.",
+          "timestamp": [
+            76.0,
+            77.14
+          ]
+        },
+        {
+          "text": " Moving",
+          "timestamp": [
+            78.08,
+            78.36
+          ]
+        },
+        {
+          "text": " on",
+          "timestamp": [
+            78.36,
+            78.5
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            78.5,
+            78.6
+          ]
+        },
+        {
+          "text": " today's",
+          "timestamp": [
+            78.6,
+            78.94
+          ]
+        },
+        {
+          "text": " agenda,",
+          "timestamp": [
+            78.94,
+            79.56
+          ]
+        },
+        {
+          "text": " I'll",
+          "timestamp": [
+            79.56,
+            79.9
+          ]
+        },
+        {
+          "text": " make",
+          "timestamp": [
+            79.9,
+            80.02
+          ]
+        },
+        {
+          "text": " some",
+          "timestamp": [
+            80.02,
+            80.18
+          ]
+        },
+        {
+          "text": " opening",
+          "timestamp": [
+            80.18,
+            80.48
+          ]
+        },
+        {
+          "text": " remarks.",
+          "timestamp": [
+            80.48,
+            81.12
+          ]
+        },
+        {
+          "text": " Devang",
+          "timestamp": [
+            81.42,
+            81.74
+          ]
+        },
+        {
+          "text": " will",
+          "timestamp": [
+            81.74,
+            81.92
+          ]
+        },
+        {
+          "text": " cover",
+          "timestamp": [
+            81.92,
+            82.16
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            82.16,
+            82.3
+          ]
+        },
+        {
+          "text": " numbers",
+          "timestamp": [
+            82.3,
+            82.54
+          ]
+        },
+        {
+          "text": " in",
+          "timestamp": [
+            82.54,
+            82.7
+          ]
+        },
+        {
+          "text": " detail.",
+          "timestamp": [
+            82.7,
+            83.3
+          ]
+        },
+        {
+          "text": " Pradab,",
+          "timestamp": [
+            83.58,
+            84.06
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            84.06,
+            84.1
+          ]
+        },
+        {
+          "text": " deposit",
+          "timestamp": [
+            84.1,
+            84.46
+          ]
+        },
+        {
+          "text": " franchise,",
+          "timestamp": [
+            84.46,
+            85.3
+          ]
+        },
+        {
+          "text": " Yomesh,",
+          "timestamp": [
+            85.3,
+            85.78
+          ]
+        },
+        {
+          "text": " consumer",
+          "timestamp": [
+            85.78,
+            86.04
+          ]
+        },
+        {
+          "text": " assets,",
+          "timestamp": [
+            86.04,
+            87.14
+          ]
+        },
+        {
+          "text": " Manish,",
+          "timestamp": [
+            87.14,
+            87.6
+          ]
+        },
+        {
+          "text": " commercial",
+          "timestamp": [
+            87.6,
+            87.76
+          ]
+        },
+        {
+          "text": " banking,",
+          "timestamp": [
+            87.76,
+            88.86
+          ]
+        },
+        {
+          "text": " Paritosh,",
+          "timestamp": [
+            88.86,
+            89.56
+          ]
+        },
+        {
+          "text": " wholesale",
+          "timestamp": [
+            89.56,
+            89.84
+          ]
+        },
+        {
+          "text": " banking",
+          "timestamp": [
+            89.84,
+            90.24
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            90.24,
+            90.46
+          ]
+        },
+        {
+          "text": " institutional",
+          "timestamp": [
+            90.46,
+            90.88
+          ]
+        },
+        {
+          "text": " businesses,",
+          "timestamp": [
+            90.88,
+            91.68
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            91.68,
+            92.1
+          ]
+        },
+        {
+          "text": " finally,",
+          "timestamp": [
+            92.1,
+            92.6
+          ]
+        },
+        {
+          "text": " Jerry",
+          "timestamp": [
+            92.6,
+            92.74
+          ]
+        },
+        {
+          "text": " will",
+          "timestamp": [
+            92.74,
+            93.44
+          ]
+        },
+        {
+          "text": " update",
+          "timestamp": [
+            93.44,
+            93.84
+          ]
+        },
+        {
+          "text": " you",
+          "timestamp": [
+            93.84,
+            94.12
+          ]
+        },
+        {
+          "text": " on",
+          "timestamp": [
+            94.12,
+            94.44
+          ]
+        },
+        {
+          "text": " our",
+          "timestamp": [
+            94.44,
+            94.66
+          ]
+        },
+        {
+          "text": " subs.",
+          "timestamp": [
+            94.66,
+            95.78
+          ]
+        },
+        {
+          "text": " Let",
+          "timestamp": [
+            96.54,
+            96.74
+          ]
+        },
+        {
+          "text": " me",
+          "timestamp": [
+            96.74,
+            96.86
+          ]
+        },
+        {
+          "text": " first",
+          "timestamp": [
+            96.86,
+            97.12
+          ]
+        },
+        {
+          "text": " talk",
+          "timestamp": [
+            97.12,
+            97.34
+          ]
+        },
+        {
+          "text": " a",
+          "timestamp": [
+            97.34,
+            97.46
+          ]
+        },
+        {
+          "text": " little",
+          "timestamp": [
+            97.46,
+            97.6
+          ]
+        },
+        {
+          "text": " bit",
+          "timestamp": [
+            97.6,
+            97.74
+          ]
+        },
+        {
+          "text": " about",
+          "timestamp": [
+            97.74,
+            98.02
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            98.02,
+            98.32
+          ]
+        },
+        {
+          "text": " macroeconomic",
+          "timestamp": [
+            98.32,
+            99.38
+          ]
+        },
+        {
+          "text": " environment.",
+          "timestamp": [
+            99.38,
+            100.24
+          ]
+        },
+        {
+          "text": " The",
+          "timestamp": [
+            101.2,
+            101.4
+          ]
+        },
+        {
+          "text": " global",
+          "timestamp": [
+            101.4,
+            101.68
+          ]
+        },
+        {
+          "text": " landscape",
+          "timestamp": [
+            101.68,
+            102.12
+          ]
+        },
+        {
+          "text": " has",
+          "timestamp": [
+            102.12,
+            102.56
+          ]
+        },
+        {
+          "text": " become",
+          "timestamp": [
+            102.56,
+            102.82
+          ]
+        },
+        {
+          "text": " more",
+          "timestamp": [
+            102.82,
+            103.06
+          ]
+        },
+        {
+          "text": " volatile.",
+          "timestamp": [
+            103.06,
+            103.84
+          ]
+        },
+        {
+          "text": " Geopolitical",
+          "timestamp": [
+            104.12,
+            104.84
+          ]
+        },
+        {
+          "text": " tensions,",
+          "timestamp": [
+            104.84,
+            105.62
+          ]
+        },
+        {
+          "text": " trade",
+          "timestamp": [
+            105.62,
+            105.88
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            105.88,
+            106.04
+          ]
+        },
+        {
+          "text": " tariff",
+          "timestamp": [
+            106.04,
+            106.38
+          ]
+        },
+        {
+          "text": " uncertainties",
+          "timestamp": [
+            106.38,
+            107.14
+          ]
+        },
+        {
+          "text": " have",
+          "timestamp": [
+            107.14,
+            107.76
+          ]
+        },
+        {
+          "text": " weighed",
+          "timestamp": [
+            107.76,
+            108.06
+          ]
+        },
+        {
+          "text": " on",
+          "timestamp": [
+            108.06,
+            108.4
+          ]
+        },
+        {
+          "text": " global",
+          "timestamp": [
+            108.4,
+            108.8
+          ]
+        },
+        {
+          "text": " investor",
+          "timestamp": [
+            108.8,
+            109.24
+          ]
+        },
+        {
+          "text": " confidence.",
+          "timestamp": [
+            109.24,
+            110.12
+          ]
+        },
+        {
+          "text": " While",
+          "timestamp": [
+            110.52,
+            110.88
+          ]
+        },
+        {
+          "text": " there",
+          "timestamp": [
+            110.88,
+            111.02
+          ]
+        },
+        {
+          "text": " seems",
+          "timestamp": [
+            111.02,
+            111.28
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            111.28,
+            111.42
+          ]
+        },
+        {
+          "text": " be",
+          "timestamp": [
+            111.42,
+            111.5
+          ]
+        },
+        {
+          "text": " a",
+          "timestamp": [
+            111.5,
+            111.6
+          ]
+        },
+        {
+          "text": " high",
+          "timestamp": [
+            111.6,
+            111.78
+          ]
+        },
+        {
+          "text": " level",
+          "timestamp": [
+            111.78,
+            112.02
+          ]
+        },
+        {
+          "text": " of",
+          "timestamp": [
+            112.02,
+            112.16
+          ]
+        },
+        {
+          "text": " confidence",
+          "timestamp": [
+            112.16,
+            112.58
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            112.58,
+            112.84
+          ]
+        },
+        {
+          "text": " get",
+          "timestamp": [
+            112.84,
+            112.98
+          ]
+        },
+        {
+          "text": " a",
+          "timestamp": [
+            112.98,
+            113.08
+          ]
+        },
+        {
+          "text": " trade",
+          "timestamp": [
+            113.08,
+            113.32
+          ]
+        },
+        {
+          "text": " deal",
+          "timestamp": [
+            113.32,
+            113.6
+          ]
+        },
+        {
+          "text": " with",
+          "timestamp": [
+            113.6,
+            113.78
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            113.78,
+            113.9
+          ]
+        },
+        {
+          "text": " EU,",
+          "timestamp": [
+            113.9,
+            114.46
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            114.82,
+            114.98
+          ]
+        },
+        {
+          "text": " absence",
+          "timestamp": [
+            114.98,
+            115.36
+          ]
+        },
+        {
+          "text": " of",
+          "timestamp": [
+            115.36,
+            115.6
+          ]
+        },
+        {
+          "text": " a",
+          "timestamp": [
+            115.6,
+            115.74
+          ]
+        },
+        {
+          "text": " final",
+          "timestamp": [
+            115.74,
+            116.08
+          ]
+        },
+        {
+          "text": " trade",
+          "timestamp": [
+            116.08,
+            116.38
+          ]
+        },
+        {
+          "text": " agreement",
+          "timestamp": [
+            116.38,
+            116.74
+          ]
+        },
+        {
+          "text": " with",
+          "timestamp": [
+            116.74,
+            116.96
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            116.96,
+            117.08
+          ]
+        },
+        {
+          "text": " US",
+          "timestamp": [
+            117.08,
+            117.46
+          ]
+        },
+        {
+          "text": " continues",
+          "timestamp": [
+            117.46,
+            118.3
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            118.3,
+            118.68
+          ]
+        },
+        {
+          "text": " add",
+          "timestamp": [
+            118.68,
+            119.04
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            119.04,
+            119.3
+          ]
+        },
+        {
+          "text": " this",
+          "timestamp": [
+            119.3,
+            119.52
+          ]
+        },
+        {
+          "text": " uncertainty.",
+          "timestamp": [
+            119.52,
+            120.34
+          ]
+        },
+        {
+          "text": " We",
+          "timestamp": [
+            121.08,
+            121.28
+          ]
+        },
+        {
+          "text": " have",
+          "timestamp": [
+            121.28,
+            121.38
+          ]
+        },
+        {
+          "text": " seen",
+          "timestamp": [
+            121.38,
+            121.6
+          ]
+        },
+        {
+          "text": " unprecedented",
+          "timestamp": [
+            121.6,
+            122.26
+          ]
+        },
+        {
+          "text": " strengthening",
+          "timestamp": [
+            122.26,
+            122.84
+          ]
+        },
+        {
+          "text": " of",
+          "timestamp": [
+            122.84,
+            123.08
+          ]
+        },
+        {
+          "text": " gold",
+          "timestamp": [
+            123.08,
+            123.34
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            123.34,
+            123.48
+          ]
+        },
+        {
+          "text": " silver",
+          "timestamp": [
+            123.48,
+            123.76
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            123.76,
+            124.6
+          ]
+        },
+        {
+          "text": " outflows",
+          "timestamp": [
+            124.6,
+            125.22
+          ]
+        },
+        {
+          "text": " from",
+          "timestamp": [
+            125.22,
+            125.48
+          ]
+        },
+        {
+          "text": " India",
+          "timestamp": [
+            125.48,
+            125.76
+          ]
+        },
+        {
+          "text": " by",
+          "timestamp": [
+            125.76,
+            126.06
+          ]
+        },
+        {
+          "text": " FIIs",
+          "timestamp": [
+            126.06,
+            126.76
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            127.1,
+            127.24
+          ]
+        },
+        {
+          "text": " other",
+          "timestamp": [
+            127.24,
+            127.52
+          ]
+        },
+        {
+          "text": " markets",
+          "timestamp": [
+            127.52,
+            127.92
+          ]
+        },
+        {
+          "text": " that",
+          "timestamp": [
+            127.92,
+            128.12
+          ]
+        },
+        {
+          "text": " have",
+          "timestamp": [
+            128.12,
+            128.22
+          ]
+        },
+        {
+          "text": " perceived",
+          "timestamp": [
+            128.22,
+            128.52
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            128.52,
+            128.8
+          ]
+        },
+        {
+          "text": " be",
+          "timestamp": [
+            128.8,
+            128.92
+          ]
+        },
+        {
+          "text": " more",
+          "timestamp": [
+            128.92,
+            129.16
+          ]
+        },
+        {
+          "text": " attractive",
+          "timestamp": [
+            129.16,
+            129.64
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            129.64,
+            130.24
+          ]
+        },
+        {
+          "text": " foreign",
+          "timestamp": [
+            130.24,
+            130.56
+          ]
+        },
+        {
+          "text": " investors.",
+          "timestamp": [
+            130.56,
+            131.6
+          ]
+        },
+        {
+          "text": " However,",
+          "timestamp": [
+            132.72,
+            133.54
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            133.54,
+            133.7
+          ]
+        },
+        {
+          "text": " Indian",
+          "timestamp": [
+            133.7,
+            133.96
+          ]
+        },
+        {
+          "text": " economy",
+          "timestamp": [
+            133.96,
+            134.46
+          ]
+        },
+        {
+          "text": " continues",
+          "timestamp": [
+            134.46,
+            134.9
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            134.9,
+            135.12
+          ]
+        },
+        {
+          "text": " demonstrate",
+          "timestamp": [
+            135.12,
+            135.66
+          ]
+        },
+        {
+          "text": " strong",
+          "timestamp": [
+            135.66,
+            136.5
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            136.5,
+            136.8
+          ]
+        },
+        {
+          "text": " stable",
+          "timestamp": [
+            136.8,
+            137.18
+          ]
+        },
+        {
+          "text": " growth.",
+          "timestamp": [
+            137.18,
+            137.74
+          ]
+        },
+        {
+          "text": " GDP",
+          "timestamp": [
+            138.32,
+            138.7
+          ]
+        },
+        {
+          "text": " growth",
+          "timestamp": [
+            138.7,
+            139.12
+          ]
+        },
+        {
+          "text": " continues",
+          "timestamp": [
+            139.12,
+            139.7
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            139.7,
+            139.88
+          ]
+        },
+        {
+          "text": " be",
+          "timestamp": [
+            139.88,
+            139.98
+          ]
+        },
+        {
+          "text": " resilient",
+          "timestamp": [
+            139.98,
+            140.38
+          ]
+        },
+        {
+          "text": " with",
+          "timestamp": [
+            140.38,
+            140.96
+          ]
+        },
+        {
+          "text": " a",
+          "timestamp": [
+            140.96,
+            141.06
+          ]
+        },
+        {
+          "text": " positive",
+          "timestamp": [
+            141.06,
+            141.38
+          ]
+        },
+        {
+          "text": " outlook.",
+          "timestamp": [
+            141.38,
+            142.16
+          ]
+        },
+        {
+          "text": " This",
+          "timestamp": [
+            142.66,
+            142.84
+          ]
+        },
+        {
+          "text": " is",
+          "timestamp": [
+            142.84,
+            142.96
+          ]
+        },
+        {
+          "text": " supported",
+          "timestamp": [
+            142.96,
+            143.34
+          ]
+        },
+        {
+          "text": " by",
+          "timestamp": [
+            143.34,
+            143.56
+          ]
+        },
+        {
+          "text": " several",
+          "timestamp": [
+            143.56,
+            143.98
+          ]
+        },
+        {
+          "text": " enabling",
+          "timestamp": [
+            143.98,
+            144.46
+          ]
+        },
+        {
+          "text": " actions",
+          "timestamp": [
+            144.46,
+            144.94
+          ]
+        },
+        {
+          "text": " like",
+          "timestamp": [
+            144.94,
+            145.76
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            145.76,
+            145.96
+          ]
+        },
+        {
+          "text": " RBI's",
+          "timestamp": [
+            145.96,
+            146.46
+          ]
+        },
+        {
+          "text": " repo",
+          "timestamp": [
+            146.46,
+            146.68
+          ]
+        },
+        {
+          "text": " rate",
+          "timestamp": [
+            146.68,
+            146.9
+          ]
+        },
+        {
+          "text": " cuts",
+          "timestamp": [
+            146.9,
+            147.3
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            147.3,
+            147.92
+          ]
+        },
+        {
+          "text": " GST",
+          "timestamp": [
+            147.92,
+            148.44
+          ]
+        },
+        {
+          "text": " rate",
+          "timestamp": [
+            148.44,
+            149.06
+          ]
+        },
+        {
+          "text": " rationalization.",
+          "timestamp": [
+            149.06,
+            150.28
+          ]
+        },
+        {
+          "text": " 3Q",
+          "timestamp": [
+            151.06,
+            151.48
+          ]
+        },
+        {
+          "text": " also",
+          "timestamp": [
+            151.48,
+            151.98
+          ]
+        },
+        {
+          "text": " saw",
+          "timestamp": [
+            151.98,
+            152.32
+          ]
+        },
+        {
+          "text": " an",
+          "timestamp": [
+            152.32,
+            152.62
+          ]
+        },
+        {
+          "text": " uptick",
+          "timestamp": [
+            152.62,
+            153.02
+          ]
+        },
+        {
+          "text": " in",
+          "timestamp": [
+            153.02,
+            153.12
+          ]
+        },
+        {
+          "text": " growth",
+          "timestamp": [
+            153.12,
+            153.44
+          ]
+        },
+        {
+          "text": " led",
+          "timestamp": [
+            153.44,
+            154.0
+          ]
+        },
+        {
+          "text": " by",
+          "timestamp": [
+            154.0,
+            154.18
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            154.18,
+            154.32
+          ]
+        },
+        {
+          "text": " usual",
+          "timestamp": [
+            154.32,
+            154.72
+          ]
+        },
+        {
+          "text": " festive",
+          "timestamp": [
+            154.72,
+            155.18
+          ]
+        },
+        {
+          "text": " demand.",
+          "timestamp": [
+            155.18,
+            156.16
+          ]
+        },
+        {
+          "text": " Rural",
+          "timestamp": [
+            156.76,
+            157.1
+          ]
+        },
+        {
+          "text": " India",
+          "timestamp": [
+            157.1,
+            157.44
+          ]
+        },
+        {
+          "text": " overall",
+          "timestamp": [
+            157.44,
+            157.96
+          ]
+        },
+        {
+          "text": " looks",
+          "timestamp": [
+            157.96,
+            158.26
+          ]
+        },
+        {
+          "text": " stable,",
+          "timestamp": [
+            158.26,
+            158.78
+          ]
+        },
+        {
+          "text": " supported",
+          "timestamp": [
+            158.78,
+            159.26
+          ]
+        },
+        {
+          "text": " by",
+          "timestamp": [
+            159.26,
+            159.54
+          ]
+        },
+        {
+          "text": " good",
+          "timestamp": [
+            159.54,
+            159.84
+          ]
+        },
+        {
+          "text": " monsuits.",
+          "timestamp": [
+            159.84,
+            161.14
+          ]
+        },
+        {
+          "text": " The",
+          "timestamp": [
+            161.52,
+            161.8
+          ]
+        },
+        {
+          "text": " rupee",
+          "timestamp": [
+            161.8,
+            162.06
+          ]
+        },
+        {
+          "text": " weakness",
+          "timestamp": [
+            162.06,
+            162.5
+          ]
+        },
+        {
+          "text": " has",
+          "timestamp": [
+            162.5,
+            162.84
+          ]
+        },
+        {
+          "text": " partly",
+          "timestamp": [
+            162.84,
+            163.16
+          ]
+        },
+        {
+          "text": " offset",
+          "timestamp": [
+            163.16,
+            163.72
+          ]
+        },
+        {
+          "text": " tariff",
+          "timestamp": [
+            163.72,
+            164.16
+          ]
+        },
+        {
+          "text": " impacts,",
+          "timestamp": [
+            164.16,
+            164.82
+          ]
+        },
+        {
+          "text": " but",
+          "timestamp": [
+            164.82,
+            165.24
+          ]
+        },
+        {
+          "text": " has",
+          "timestamp": [
+            165.24,
+            165.4
+          ]
+        },
+        {
+          "text": " also",
+          "timestamp": [
+            165.4,
+            165.8
+          ]
+        },
+        {
+          "text": " affected",
+          "timestamp": [
+            165.8,
+            166.26
+          ]
+        },
+        {
+          "text": " foreign",
+          "timestamp": [
+            166.26,
+            166.62
+          ]
+        },
+        {
+          "text": " flows",
+          "timestamp": [
+            166.62,
+            167.06
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            167.06,
+            167.76
+          ]
+        },
+        {
+          "text": " liquidity.",
+          "timestamp": [
+            167.76,
+            168.6
+          ]
+        },
+        {
+          "text": " With",
+          "timestamp": [
+            169.76,
+            169.98
+          ]
+        },
+        {
+          "text": " respect",
+          "timestamp": [
+            169.98,
+            170.26
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            170.26,
+            170.42
+          ]
+        },
+        {
+          "text": " capital",
+          "timestamp": [
+            170.42,
+            170.76
+          ]
+        },
+        {
+          "text": " flows,",
+          "timestamp": [
+            170.76,
+            171.22
+          ]
+        },
+        {
+          "text": " while",
+          "timestamp": [
+            171.22,
+            171.44
+          ]
+        },
+        {
+          "text": " FIIs",
+          "timestamp": [
+            171.44,
+            172.02
+          ]
+        },
+        {
+          "text": " have",
+          "timestamp": [
+            172.02,
+            172.14
+          ]
+        },
+        {
+          "text": " been",
+          "timestamp": [
+            172.14,
+            172.28
+          ]
+        },
+        {
+          "text": " net",
+          "timestamp": [
+            172.28,
+            172.5
+          ]
+        },
+        {
+          "text": " sellers,",
+          "timestamp": [
+            172.5,
+            173.4
+          ]
+        },
+        {
+          "text": " domestic",
+          "timestamp": [
+            173.48,
+            173.82
+          ]
+        },
+        {
+          "text": " institutional",
+          "timestamp": [
+            173.82,
+            175.02
+          ]
+        },
+        {
+          "text": " investors",
+          "timestamp": [
+            175.02,
+            175.58
+          ]
+        },
+        {
+          "text": " have",
+          "timestamp": [
+            175.58,
+            176.4
+          ]
+        },
+        {
+          "text": " provided",
+          "timestamp": [
+            176.4,
+            176.74
+          ]
+        },
+        {
+          "text": " support",
+          "timestamp": [
+            176.74,
+            177.18
+          ]
+        },
+        {
+          "text": " with",
+          "timestamp": [
+            177.18,
+            177.48
+          ]
+        },
+        {
+          "text": " around",
+          "timestamp": [
+            177.48,
+            177.74
+          ]
+        },
+        {
+          "text": " $24",
+          "timestamp": [
+            177.74,
+            178.32
+          ]
+        },
+        {
+          "text": " billion",
+          "timestamp": [
+            178.32,
+            178.92
+          ]
+        },
+        {
+          "text": " in",
+          "timestamp": [
+            178.92,
+            180.18
+          ]
+        },
+        {
+          "text": " inflows",
+          "timestamp": [
+            180.18,
+            180.82
+          ]
+        },
+        {
+          "text": " in",
+          "timestamp": [
+            180.82,
+            181.36
+          ]
+        },
+        {
+          "text": " Q3",
+          "timestamp": [
+            181.36,
+            181.86
+          ]
+        },
+        {
+          "text": " FY26,",
+          "timestamp": [
+            181.86,
+            183.42
+          ]
+        },
+        {
+          "text": " reinforcing",
+          "timestamp": [
+            183.84,
+            184.52
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            184.52,
+            184.84
+          ]
+        },
+        {
+          "text": " savers",
+          "timestamp": [
+            184.84,
+            185.18
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            185.18,
+            185.34
+          ]
+        },
+        {
+          "text": " investors'",
+          "timestamp": [
+            185.34,
+            186.04
+          ]
+        },
+        {
+          "text": " strength.",
+          "timestamp": [
+            186.04,
+            186.42
+          ]
+        },
+        {
+          "text": " This",
+          "timestamp": [
+            186.36,
+            187.34
+          ]
+        },
+        {
+          "text": " is",
+          "timestamp": [
+            187.34,
+            187.48
+          ]
+        },
+        {
+          "text": " also",
+          "timestamp": [
+            187.48,
+            187.86
+          ]
+        },
+        {
+          "text": " evident",
+          "timestamp": [
+            187.86,
+            188.28
+          ]
+        },
+        {
+          "text": " in",
+          "timestamp": [
+            188.28,
+            188.58
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            188.58,
+            188.68
+          ]
+        },
+        {
+          "text": " strong",
+          "timestamp": [
+            188.68,
+            189.04
+          ]
+        },
+        {
+          "text": " IPO",
+          "timestamp": [
+            189.04,
+            189.54
+          ]
+        },
+        {
+          "text": " pipeline",
+          "timestamp": [
+            189.54,
+            190.08
+          ]
+        },
+        {
+          "text": " expected",
+          "timestamp": [
+            190.08,
+            191.22
+          ]
+        },
+        {
+          "text": " for",
+          "timestamp": [
+            191.22,
+            191.58
+          ]
+        },
+        {
+          "text": " calendar",
+          "timestamp": [
+            191.58,
+            191.94
+          ]
+        },
+        {
+          "text": " 2026.",
+          "timestamp": [
+            191.94,
+            193.68
+          ]
+        },
+        {
+          "text": " In",
+          "timestamp": [
+            194.54,
+            194.7
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            194.7,
+            194.8
+          ]
+        },
+        {
+          "text": " banking",
+          "timestamp": [
+            194.8,
+            195.12
+          ]
+        },
+        {
+          "text": " sector",
+          "timestamp": [
+            195.12,
+            195.54
+          ]
+        },
+        {
+          "text": " specifically,",
+          "timestamp": [
+            195.54,
+            196.74
+          ]
+        },
+        {
+          "text": " system",
+          "timestamp": [
+            196.74,
+            197.08
+          ]
+        },
+        {
+          "text": " credit",
+          "timestamp": [
+            197.08,
+            197.42
+          ]
+        },
+        {
+          "text": " growth",
+          "timestamp": [
+            197.42,
+            197.76
+          ]
+        },
+        {
+          "text": " has",
+          "timestamp": [
+            197.76,
+            197.92
+          ]
+        },
+        {
+          "text": " been",
+          "timestamp": [
+            197.92,
+            198.06
+          ]
+        },
+        {
+          "text": " healthy,",
+          "timestamp": [
+            198.06,
+            198.62
+          ]
+        },
+        {
+          "text": " but",
+          "timestamp": [
+            199.02,
+            199.16
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            199.16,
+            199.26
+          ]
+        },
+        {
+          "text": " flows",
+          "timestamp": [
+            199.26,
+            199.52
+          ]
+        },
+        {
+          "text": " into",
+          "timestamp": [
+            199.52,
+            199.76
+          ]
+        },
+        {
+          "text": " commodities",
+          "timestamp": [
+            199.76,
+            200.26
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            200.26,
+            200.5
+          ]
+        },
+        {
+          "text": " capital",
+          "timestamp": [
+            200.5,
+            200.84
+          ]
+        },
+        {
+          "text": " markets",
+          "timestamp": [
+            200.84,
+            201.34
+          ]
+        },
+        {
+          "text": " are",
+          "timestamp": [
+            201.34,
+            201.92
+          ]
+        },
+        {
+          "text": " putting",
+          "timestamp": [
+            201.92,
+            202.16
+          ]
+        },
+        {
+          "text": " more",
+          "timestamp": [
+            202.16,
+            202.48
+          ]
+        },
+        {
+          "text": " pressure",
+          "timestamp": [
+            202.48,
+            202.88
+          ]
+        },
+        {
+          "text": " on",
+          "timestamp": [
+            202.88,
+            203.46
+          ]
+        },
+        {
+          "text": " low",
+          "timestamp": [
+            203.46,
+            203.7
+          ]
+        },
+        {
+          "text": "-cost",
+          "timestamp": [
+            203.7,
+            203.98
+          ]
+        },
+        {
+          "text": " deposits.",
+          "timestamp": [
+            203.98,
+            204.82
+          ]
+        },
+        {
+          "text": " There",
+          "timestamp": [
+            205.42,
+            205.64
+          ]
+        },
+        {
+          "text": " is",
+          "timestamp": [
+            205.64,
+            205.78
+          ]
+        },
+        {
+          "text": " enhanced",
+          "timestamp": [
+            205.78,
+            206.3
+          ]
+        },
+        {
+          "text": " volatility",
+          "timestamp": [
+            206.3,
+            206.8
+          ]
+        },
+        {
+          "text": " seen",
+          "timestamp": [
+            206.8,
+            207.64
+          ]
+        },
+        {
+          "text": " in",
+          "timestamp": [
+            207.64,
+            207.8
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            207.8,
+            207.88
+          ]
+        },
+        {
+          "text": " banking",
+          "timestamp": [
+            207.88,
+            208.16
+          ]
+        },
+        {
+          "text": " sector",
+          "timestamp": [
+            208.16,
+            208.52
+          ]
+        },
+        {
+          "text": " liquidity.",
+          "timestamp": [
+            208.52,
+            209.34
+          ]
+        },
+        {
+          "text": " While",
+          "timestamp": [
+            209.7,
+            209.9
+          ]
+        },
+        {
+          "text": " benchmark",
+          "timestamp": [
+            209.9,
+            210.28
+          ]
+        },
+        {
+          "text": " rates",
+          "timestamp": [
+            210.28,
+            210.68
+          ]
+        },
+        {
+          "text": " are",
+          "timestamp": [
+            210.68,
+            210.84
+          ]
+        },
+        {
+          "text": " reduced,",
+          "timestamp": [
+            210.84,
+            211.8
+          ]
+        },
+        {
+          "text": " longer",
+          "timestamp": [
+            211.8,
+            212.14
+          ]
+        },
+        {
+          "text": "-term",
+          "timestamp": [
+            212.14,
+            212.54
+          ]
+        },
+        {
+          "text": " credit",
+          "timestamp": [
+            212.54,
+            212.98
+          ]
+        },
+        {
+          "text": " yields",
+          "timestamp": [
+            212.98,
+            213.44
+          ]
+        },
+        {
+          "text": " have",
+          "timestamp": [
+            213.44,
+            213.94
+          ]
+        },
+        {
+          "text": " gone",
+          "timestamp": [
+            213.94,
+            214.24
+          ]
+        },
+        {
+          "text": " down.",
+          "timestamp": [
+            214.24,
+            214.84
+          ]
+        },
+        {
+          "text": " Within",
+          "timestamp": [
+            216.26,
+            216.58
+          ]
+        },
+        {
+          "text": " this",
+          "timestamp": [
+            216.58,
+            216.82
+          ]
+        },
+        {
+          "text": " context,",
+          "timestamp": [
+            216.82,
+            217.54
+          ]
+        },
+        {
+          "text": " coming",
+          "timestamp": [
+            217.54,
+            217.78
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            217.78,
+            217.94
+          ]
+        },
+        {
+          "text": " quota,",
+          "timestamp": [
+            217.94,
+            218.48
+          ]
+        },
+        {
+          "text": " we",
+          "timestamp": [
+            218.48,
+            218.7
+          ]
+        },
+        {
+          "text": " remain",
+          "timestamp": [
+            218.7,
+            218.98
+          ]
+        },
+        {
+          "text": " focused",
+          "timestamp": [
+            218.98,
+            219.38
+          ]
+        },
+        {
+          "text": " on",
+          "timestamp": [
+            219.38,
+            219.6
+          ]
+        },
+        {
+          "text": " our",
+          "timestamp": [
+            219.6,
+            219.78
+          ]
+        },
+        {
+          "text": " strategy",
+          "timestamp": [
+            219.78,
+            220.18
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            220.18,
+            220.8
+          ]
+        },
+        {
+          "text": " scale",
+          "timestamp": [
+            220.8,
+            221.24
+          ]
+        },
+        {
+          "text": " responsibly.",
+          "timestamp": [
+            221.24,
+            222.42
+          ]
+        },
+        {
+          "text": " This",
+          "timestamp": [
+            222.98,
+            223.18
+          ]
+        },
+        {
+          "text": " discipline",
+          "timestamp": [
+            223.18,
+            223.6
+          ]
+        },
+        {
+          "text": " is",
+          "timestamp": [
+            223.6,
+            223.96
+          ]
+        },
+        {
+          "text": " reflected",
+          "timestamp": [
+            223.96,
+            224.38
+          ]
+        },
+        {
+          "text": " in",
+          "timestamp": [
+            224.38,
+            224.78
+          ]
+        },
+        {
+          "text": " our",
+          "timestamp": [
+            224.78,
+            224.88
+          ]
+        },
+        {
+          "text": " quarterly",
+          "timestamp": [
+            224.88,
+            225.22
+          ]
+        },
+        {
+          "text": " performance.",
+          "timestamp": [
+            225.22,
+            226.14
+          ]
+        },
+        {
+          "text": " In",
+          "timestamp": [
+            227.04,
+            227.24
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            227.24,
+            227.34
+          ]
+        },
+        {
+          "text": " bank,",
+          "timestamp": [
+            227.34,
+            227.94
+          ]
+        },
+        {
+          "text": " net",
+          "timestamp": [
+            227.94,
+            228.08
+          ]
+        },
+        {
+          "text": " advances",
+          "timestamp": [
+            228.08,
+            228.54
+          ]
+        },
+        {
+          "text": " grew",
+          "timestamp": [
+            228.54,
+            228.86
+          ]
+        },
+        {
+          "text": " 16",
+          "timestamp": [
+            228.86,
+            229.26
+          ]
+        },
+        {
+          "text": "%",
+          "timestamp": [
+            229.26,
+            229.78
+          ]
+        },
+        {
+          "text": " YOY.",
+          "timestamp": [
+            229.78,
+            231.16
+          ]
+        },
+        {
+          "text": " This",
+          "timestamp": [
+            231.64,
+            231.78
+          ]
+        },
+        {
+          "text": " is",
+          "timestamp": [
+            231.78,
+            231.9
+          ]
+        },
+        {
+          "text": " in",
+          "timestamp": [
+            231.9,
+            232.1
+          ]
+        },
+        {
+          "text": " line",
+          "timestamp": [
+            232.1,
+            232.22
+          ]
+        },
+        {
+          "text": " with",
+          "timestamp": [
+            232.22,
+            232.42
+          ]
+        },
+        {
+          "text": " our",
+          "timestamp": [
+            232.42,
+            232.56
+          ]
+        },
+        {
+          "text": " stated",
+          "timestamp": [
+            232.56,
+            232.82
+          ]
+        },
+        {
+          "text": " philosophy",
+          "timestamp": [
+            232.82,
+            233.36
+          ]
+        },
+        {
+          "text": " of",
+          "timestamp": [
+            233.36,
+            234.08
+          ]
+        },
+        {
+          "text": " growing",
+          "timestamp": [
+            234.08,
+            234.48
+          ]
+        },
+        {
+          "text": " our",
+          "timestamp": [
+            234.48,
+            234.76
+          ]
+        },
+        {
+          "text": " advances",
+          "timestamp": [
+            234.76,
+            235.2
+          ]
+        },
+        {
+          "text": " in",
+          "timestamp": [
+            235.2,
+            235.94
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            235.94,
+            236.06
+          ]
+        },
+        {
+          "text": " range",
+          "timestamp": [
+            236.06,
+            236.34
+          ]
+        },
+        {
+          "text": " of",
+          "timestamp": [
+            236.34,
+            236.52
+          ]
+        },
+        {
+          "text": " 1",
+          "timestamp": [
+            236.52,
+            236.68
+          ]
+        },
+        {
+          "text": ".5",
+          "timestamp": [
+            236.68,
+            237.02
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            237.34,
+            237.52
+          ]
+        },
+        {
+          "text": " 2",
+          "timestamp": [
+            237.52,
+            237.72
+          ]
+        },
+        {
+          "text": " times",
+          "timestamp": [
+            237.72,
+            238.1
+          ]
+        },
+        {
+          "text": " nominal",
+          "timestamp": [
+            238.1,
+            238.52
+          ]
+        },
+        {
+          "text": " GDP",
+          "timestamp": [
+            238.52,
+            238.9
+          ]
+        },
+        {
+          "text": " growth.",
+          "timestamp": [
+            238.9,
+            239.52
+          ]
+        },
+        {
+          "text": " We",
+          "timestamp": [
+            239.34,
+            240.4
+          ]
+        },
+        {
+          "text": " have",
+          "timestamp": [
+            240.4,
+            240.52
+          ]
+        },
+        {
+          "text": " seen",
+          "timestamp": [
+            240.52,
+            240.72
+          ]
+        },
+        {
+          "text": " consistent",
+          "timestamp": [
+            240.72,
+            241.24
+          ]
+        },
+        {
+          "text": " growth",
+          "timestamp": [
+            241.24,
+            241.58
+          ]
+        },
+        {
+          "text": " of",
+          "timestamp": [
+            241.58,
+            241.9
+          ]
+        },
+        {
+          "text": " around",
+          "timestamp": [
+            241.9,
+            242.14
+          ]
+        },
+        {
+          "text": " 4",
+          "timestamp": [
+            242.14,
+            242.38
+          ]
+        },
+        {
+          "text": "%",
+          "timestamp": [
+            242.38,
+            242.64
+          ]
+        },
+        {
+          "text": " per",
+          "timestamp": [
+            242.64,
+            242.88
+          ]
+        },
+        {
+          "text": " quarter",
+          "timestamp": [
+            242.88,
+            243.3
+          ]
+        },
+        {
+          "text": " for",
+          "timestamp": [
+            243.3,
+            243.8
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            243.8,
+            243.88
+          ]
+        },
+        {
+          "text": " last",
+          "timestamp": [
+            243.88,
+            244.16
+          ]
+        },
+        {
+          "text": " three",
+          "timestamp": [
+            244.16,
+            244.38
+          ]
+        },
+        {
+          "text": " quarters",
+          "timestamp": [
+            244.38,
+            244.74
+          ]
+        },
+        {
+          "text": " of",
+          "timestamp": [
+            244.74,
+            245.38
+          ]
+        },
+        {
+          "text": " this",
+          "timestamp": [
+            245.38,
+            245.6
+          ]
+        },
+        {
+          "text": " financial",
+          "timestamp": [
+            245.6,
+            246.02
+          ]
+        },
+        {
+          "text": " year.",
+          "timestamp": [
+            246.02,
+            246.98
+          ]
+        },
+        {
+          "text": " Having",
+          "timestamp": [
+            248.0,
+            248.28
+          ]
+        },
+        {
+          "text": " deposits",
+          "timestamp": [
+            248.28,
+            248.78
+          ]
+        },
+        {
+          "text": " grew",
+          "timestamp": [
+            248.78,
+            248.98
+          ]
+        },
+        {
+          "text": " 15",
+          "timestamp": [
+            248.98,
+            249.3
+          ]
+        },
+        {
+          "text": "%",
+          "timestamp": [
+            249.3,
+            249.64
+          ]
+        },
+        {
+          "text": " year",
+          "timestamp": [
+            249.64,
+            249.9
+          ]
+        },
+        {
+          "text": " on",
+          "timestamp": [
+            249.9,
+            250.06
+          ]
+        },
+        {
+          "text": " year,",
+          "timestamp": [
+            250.06,
+            250.58
+          ]
+        },
+        {
+          "text": " we",
+          "timestamp": [
+            250.58,
+            251.06
+          ]
+        },
+        {
+          "text": " have",
+          "timestamp": [
+            251.06,
+            251.14
+          ]
+        },
+        {
+          "text": " been",
+          "timestamp": [
+            251.14,
+            251.28
+          ]
+        },
+        {
+          "text": " focusing",
+          "timestamp": [
+            251.28,
+            251.66
+          ]
+        },
+        {
+          "text": " on",
+          "timestamp": [
+            251.66,
+            251.88
+          ]
+        },
+        {
+          "text": " granular",
+          "timestamp": [
+            251.88,
+            252.28
+          ]
+        },
+        {
+          "text": " casa",
+          "timestamp": [
+            252.28,
+            252.58
+          ]
+        },
+        {
+          "text": " growth.",
+          "timestamp": [
+            252.58,
+            253.14
+          ]
+        },
+        {
+          "text": " This",
+          "timestamp": [
+            253.62,
+            253.78
+          ]
+        },
+        {
+          "text": " is",
+          "timestamp": [
+            253.78,
+            253.9
+          ]
+        },
+        {
+          "text": " on",
+          "timestamp": [
+            253.9,
+            254.04
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            254.04,
+            254.14
+          ]
+        },
+        {
+          "text": " back",
+          "timestamp": [
+            254.14,
+            254.48
+          ]
+        },
+        {
+          "text": " of",
+          "timestamp": [
+            254.48,
+            255.1
+          ]
+        },
+        {
+          "text": " our",
+          "timestamp": [
+            255.1,
+            255.26
+          ]
+        },
+        {
+          "text": " focus",
+          "timestamp": [
+            255.26,
+            255.6
+          ]
+        },
+        {
+          "text": " segment",
+          "timestamp": [
+            255.6,
+            255.98
+          ]
+        },
+        {
+          "text": " strategy,",
+          "timestamp": [
+            255.98,
+            256.58
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            256.58,
+            257.06
+          ]
+        },
+        {
+          "text": " I",
+          "timestamp": [
+            257.06,
+            257.16
+          ]
+        },
+        {
+          "text": " will",
+          "timestamp": [
+            257.16,
+            257.3
+          ]
+        },
+        {
+          "text": " come",
+          "timestamp": [
+            257.3,
+            257.5
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            257.5,
+            257.66
+          ]
+        },
+        {
+          "text": " that",
+          "timestamp": [
+            257.66,
+            257.9
+          ]
+        },
+        {
+          "text": " shortly.",
+          "timestamp": [
+            257.9,
+            258.72
+          ]
+        },
+        {
+          "text": " We",
+          "timestamp": [
+            259.52,
+            259.84
+          ]
+        },
+        {
+          "text": " called",
+          "timestamp": [
+            259.84,
+            260.14
+          ]
+        },
+        {
+          "text": " out",
+          "timestamp": [
+            260.14,
+            260.32
+          ]
+        },
+        {
+          "text": " unsecured",
+          "timestamp": [
+            260.32,
+            260.9
+          ]
+        },
+        {
+          "text": " retail",
+          "timestamp": [
+            260.9,
+            261.18
+          ]
+        },
+        {
+          "text": " loans",
+          "timestamp": [
+            261.18,
+            261.68
+          ]
+        },
+        {
+          "text": " earlier.",
+          "timestamp": [
+            261.68,
+            262.46
+          ]
+        },
+        {
+          "text": " We",
+          "timestamp": [
+            262.98,
+            263.1
+          ]
+        },
+        {
+          "text": " said",
+          "timestamp": [
+            263.1,
+            263.28
+          ]
+        },
+        {
+          "text": " we",
+          "timestamp": [
+            263.28,
+            263.42
+          ]
+        },
+        {
+          "text": " would",
+          "timestamp": [
+            263.42,
+            263.54
+          ]
+        },
+        {
+          "text": " grow",
+          "timestamp": [
+            263.54,
+            265.04
+          ]
+        },
+        {
+          "text": " gradually",
+          "timestamp": [
+            265.04,
+            265.54
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            265.54,
+            265.9
+          ]
+        },
+        {
+          "text": " responsibly.",
+          "timestamp": [
+            265.9,
+            266.84
+          ]
+        },
+        {
+          "text": " with",
+          "timestamp": [
+            266.6,
+            267.44
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            267.44,
+            267.56
+          ]
+        },
+        {
+          "text": " unsecured",
+          "timestamp": [
+            267.56,
+            268.0
+          ]
+        },
+        {
+          "text": " assets",
+          "timestamp": [
+            268.0,
+            268.38
+          ]
+        },
+        {
+          "text": " growing",
+          "timestamp": [
+            268.38,
+            268.72
+          ]
+        },
+        {
+          "text": " in",
+          "timestamp": [
+            268.72,
+            268.94
+          ]
+        },
+        {
+          "text": " absolute",
+          "timestamp": [
+            268.94,
+            269.38
+          ]
+        },
+        {
+          "text": " terms",
+          "timestamp": [
+            269.38,
+            269.84
+          ]
+        },
+        {
+          "text": " in",
+          "timestamp": [
+            269.84,
+            270.36
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            270.36,
+            270.46
+          ]
+        },
+        {
+          "text": " first",
+          "timestamp": [
+            270.46,
+            270.72
+          ]
+        },
+        {
+          "text": " instance.",
+          "timestamp": [
+            270.72,
+            271.34
+          ]
+        },
+        {
+          "text": " And",
+          "timestamp": [
+            271.76,
+            271.88
+          ]
+        },
+        {
+          "text": " that's",
+          "timestamp": [
+            271.88,
+            272.12
+          ]
+        },
+        {
+          "text": " exactly",
+          "timestamp": [
+            272.12,
+            272.62
+          ]
+        },
+        {
+          "text": " what",
+          "timestamp": [
+            272.62,
+            272.88
+          ]
+        },
+        {
+          "text": " we",
+          "timestamp": [
+            272.88,
+            273.06
+          ]
+        },
+        {
+          "text": " have",
+          "timestamp": [
+            273.06,
+            273.2
+          ]
+        },
+        {
+          "text": " done",
+          "timestamp": [
+            273.2,
+            273.48
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            273.48,
+            274.04
+          ]
+        },
+        {
+          "text": " will",
+          "timestamp": [
+            274.04,
+            274.22
+          ]
+        },
+        {
+          "text": " continue",
+          "timestamp": [
+            274.22,
+            274.58
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            274.58,
+            275.02
+          ]
+        },
+        {
+          "text": " do",
+          "timestamp": [
+            275.02,
+            275.2
+          ]
+        },
+        {
+          "text": " so.",
+          "timestamp": [
+            275.2,
+            275.7
+          ]
+        },
+        {
+          "text": " The",
+          "timestamp": [
+            276.32,
+            276.5
+          ]
+        },
+        {
+          "text": " bank's",
+          "timestamp": [
+            276.5,
+            277.16
+          ]
+        },
+        {
+          "text": " NIL",
+          "timestamp": [
+            277.16,
+            277.36
+          ]
+        },
+        {
+          "text": " continues",
+          "timestamp": [
+            277.36,
+            277.82
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            277.82,
+            278.0
+          ]
+        },
+        {
+          "text": " be",
+          "timestamp": [
+            278.0,
+            278.22
+          ]
+        },
+        {
+          "text": " healthy",
+          "timestamp": [
+            278.22,
+            278.98
+          ]
+        },
+        {
+          "text": " at",
+          "timestamp": [
+            278.98,
+            279.42
+          ]
+        },
+        {
+          "text": " 4",
+          "timestamp": [
+            279.42,
+            279.62
+          ]
+        },
+        {
+          "text": ".54%,",
+          "timestamp": [
+            279.62,
+            281.64
+          ]
+        },
+        {
+          "text": " supported",
+          "timestamp": [
+            281.64,
+            282.68
+          ]
+        },
+        {
+          "text": " by",
+          "timestamp": [
+            282.68,
+            282.94
+          ]
+        },
+        {
+          "text": " a",
+          "timestamp": [
+            282.94,
+            283.1
+          ]
+        },
+        {
+          "text": " low",
+          "timestamp": [
+            283.1,
+            283.28
+          ]
+        },
+        {
+          "text": " cost",
+          "timestamp": [
+            283.28,
+            283.58
+          ]
+        },
+        {
+          "text": " of",
+          "timestamp": [
+            283.58,
+            283.74
+          ]
+        },
+        {
+          "text": " funds.",
+          "timestamp": [
+            283.74,
+            284.54
+          ]
+        },
+        {
+          "text": " Credit",
+          "timestamp": [
+            284.9,
+            285.2
+          ]
+        },
+        {
+          "text": " costs,",
+          "timestamp": [
+            285.2,
+            285.62
+          ]
+        },
+        {
+          "text": " as",
+          "timestamp": [
+            285.62,
+            285.72
+          ]
+        },
+        {
+          "text": " indicated",
+          "timestamp": [
+            285.72,
+            286.14
+          ]
+        },
+        {
+          "text": " previously,",
+          "timestamp": [
+            286.14,
+            287.22
+          ]
+        },
+        {
+          "text": " continue",
+          "timestamp": [
+            287.22,
+            287.54
+          ]
+        },
+        {
+          "text": " the",
+          "timestamp": [
+            287.54,
+            287.78
+          ]
+        },
+        {
+          "text": " downward",
+          "timestamp": [
+            287.78,
+            288.1
+          ]
+        },
+        {
+          "text": " trend,",
+          "timestamp": [
+            288.1,
+            288.7
+          ]
+        },
+        {
+          "text": " supported",
+          "timestamp": [
+            288.94,
+            289.44
+          ]
+        },
+        {
+          "text": " by",
+          "timestamp": [
+            289.44,
+            289.84
+          ]
+        },
+        {
+          "text": " updated",
+          "timestamp": [
+            289.84,
+            290.58
+          ]
+        },
+        {
+          "text": " underwriting",
+          "timestamp": [
+            290.58,
+            291.06
+          ]
+        },
+        {
+          "text": " models",
+          "timestamp": [
+            291.06,
+            291.48
+          ]
+        },
+        {
+          "text": " and",
+          "timestamp": [
+            291.48,
+            292.04
+          ]
+        },
+        {
+          "text": " enhanced",
+          "timestamp": [
+            292.04,
+            292.46
+          ]
+        },
+        {
+          "text": " collection",
+          "timestamp": [
+            292.46,
+            292.88
+          ]
+        },
+        {
+          "text": " efforts.",
+          "timestamp": [
+            292.88,
+            293.52
+          ]
+        },
+        {
+          "text": " Credit",
+          "timestamp": [
+            323.32,
+            323.32
+          ]
+        },
+        {
+          "text": " costs",
+          "timestamp": [
+            323.32,
+            323.32
+          ]
+        },
+        {
+          "text": " are",
+          "timestamp": [
+            323.32,
+            323.32
+          ]
+        },
+        {
+          "text": " down",
+          "timestamp": [
+            323.32,
+            323.32
+          ]
+        },
+        {
+          "text": " sequentially",
+          "timestamp": [
+            323.32,
+            323.32
+          ]
+        },
+        {
+          "text": " from",
+          "timestamp": [
+            323.32,
+            323.32
+          ]
+        },
+        {
+          "text": " 93",
+          "timestamp": [
+            323.32,
+            323.32
+          ]
+        },
+        {
+          "text": " basis",
+          "timestamp": [
+            323.32,
+            323.32
+          ]
+        },
+        {
+          "text": " points",
+          "timestamp": [
+            323.32,
+            323.32
+          ]
+        },
+        {
+          "text": " in",
+          "timestamp": [
+            323.32,
+            323.32
+          ]
+        },
+        {
+          "text": " Q1",
+          "timestamp": [
+            323.32,
+            323.32
+          ]
+        },
+        {
+          "text": " to",
+          "timestamp": [
+            323.32,
+            323.32
+          ]
+        },
+        {
+          "text": " 79",
+          "timestamp": [
+            323.32,
+            323.32
+          ]
+        },
+        {
+          "text": " basis",
+          "timestamp": [
+            323.32,
+            323.32
+          ]
+        },
+        {
+          "text": " points",
+          "timestamp": [
+            323.32,
+            323.32
+          ]
+        },
+        {
+          "text": " in",
+          "timestamp": [
+            323.32,
+            323.32
+          ]
+        },
+        {
+          "text": " Q1.",
+          "timestamp": [
+            323.32,
+            323.32
+          ]
+        }
+      ]
+    }
+  },
+  "timestamp_granularity": "word"
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_171648/benchmark_stats.json

@@ -0,0 +1,42 @@
+{
+  "space": "Ratnesh-dev/transcribe-diarize",
+  "audio_file": "Q3-FY26_15min.mp3",
+  "task": "transcribe",
+  "language": null,
+  "models": [
+    "Whisper Large V3 Turbo",
+    "NVIDIA Parakeet v3"
+  ],
+  "benchmark_timing": {
+    "total_client_wall_clock_seconds": 97.6962
+  },
+  "leaderboard_by_gpu_window_seconds": [
+    {
+      "model": "NVIDIA Parakeet v3",
+      "api_name": "/transcribe_parakeet_v3",
+      "gpu_window_seconds": 4.3778,
+      "inference_seconds": 3.6975,
+      "client_wall_clock_seconds": 15.3745
+    }
+  ],
+  "summary": [
+    {
+      "model": "Whisper Large V3 Turbo",
+      "status": "error",
+      "api_name": "/transcribe_whisper_large_v3_turbo",
+      "client_wall_clock_seconds": 82.3215,
+      "gpu_window_seconds": null,
+      "inference_seconds": null,
+      "error": "GPU task aborted"
+    },
+    {
+      "model": "NVIDIA Parakeet v3",
+      "status": "ok",
+      "api_name": "/transcribe_parakeet_v3",
+      "client_wall_clock_seconds": 15.3745,
+      "gpu_window_seconds": 4.3778,
+      "inference_seconds": 3.6975,
+      "error": null
+    }
+  ]
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_171648/whisper_large_v3_turbo.json

@@ -0,0 +1,8 @@
+{
+  "model": "Whisper Large V3 Turbo",
+  "api_name": "/transcribe_whisper_large_v3_turbo",
+  "status": "error",
+  "client_wall_clock_seconds": 82.3215,
+  "effective_model_options_json": "{\"num_beams\": 1, \"temperature\": 0.0}",
+  "error": "GPU task aborted"
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_183018/benchmark_stats.json

@@ -0,0 +1,24 @@
+{
+  "space": "Ratnesh-dev/transcribe-diarize",
+  "audio_file": "Q3-FY26_5min.mp3",
+  "task": "diarize",
+  "language": null,
+  "models": [
+    "pyannote/speaker-diarization-community-1"
+  ],
+  "benchmark_timing": {
+    "total_client_wall_clock_seconds": 6.6803
+  },
+  "leaderboard_by_gpu_window_seconds": [],
+  "summary": [
+    {
+      "model": "pyannote/speaker-diarization-community-1",
+      "status": "error",
+      "api_name": "/diarize_pyannote_community_1",
+      "client_wall_clock_seconds": 6.6803,
+      "gpu_window_seconds": null,
+      "inference_seconds": null,
+      "error": "You have exceeded your GPU quota (60s requested vs. 22s left). Try again in 21:23:55"
+    }
+  ]
+}

archive/legacy_pre_rebuild_20260407_133100/benchmark_outputs/20260223_183018/pyannote_speaker-diarization-community-1.json

@@ -0,0 +1,11 @@
+{
+  "model": "pyannote/speaker-diarization-community-1",
+  "api_name": "/diarize_pyannote_community_1",
+  "status": "error",
+  "client_wall_clock_seconds": 6.6803,
+  "effective_model_options_json": "{\"long_audio_chunk_threshold_s\": 300, \"chunk_duration_s\": 180, \"chunk_overlap_s\": 8}",
+  "num_speakers": null,
+  "min_speakers": null,
+  "max_speakers": null,
+  "error": "You have exceeded your GPU quota (60s requested vs. 22s left). Try again in 21:23:55"
+}