Merge remote-tracking branch 't4/main'

Merge remote-tracking branch 'origin/main'

Update README: add YAML tags, all prize claims, Off Brand badge, Modal callout, gitignore Claude/dev folders

Upload readme_materials/tran_from_trans_produce.webp with huggingface_hub

Upload readme_materials/screenshot_dashboard.png with huggingface_hub

Upload readme_materials/screenshot_confirm_2.png with huggingface_hub

Upload readme_materials/screenshot_confirm.png with huggingface_hub

Upload readme_materials/screenshot_catalog.png with huggingface_hub

Upload readme_materials/screenshot_capture.png with huggingface_hub

Upload readme_materials/rocky italian market.png with huggingface_hub

Upload readme_materials/italian market snow.jpg with huggingface_hub

Upload readme_materials/italian market produce photo.jpeg with huggingface_hub

Upload readme_materials/italian market full photo.jpg with huggingface_hub

Upload readme_materials/tran_from_trans_produce.webp with huggingface_hub

Upload readme_materials/screenshot_dashboard.png with huggingface_hub

Upload readme_materials/screenshot_confirm_2.png with huggingface_hub

Upload readme_materials/screenshot_confirm.png with huggingface_hub

Upload readme_materials/screenshot_catalog.png with huggingface_hub

Upload readme_materials/screenshot_capture.png with huggingface_hub

Upload readme_materials/rocky italian market.png with huggingface_hub

Upload readme_materials/italian market snow.jpg with huggingface_hub

Upload readme_materials/italian market produce photo.jpeg with huggingface_hub

Upload readme_materials/italian market full photo.jpg with huggingface_hub

Add full README with story, architecture, and app screenshots

Compute order_total in Python from line_totals instead of trusting LLM arithmetic

Add Section 13: UI work plan (custom HTML/CSS within Gradio)

Update context docs: T4 persistent GPU, deployment strategy, Codex final submission reminder

Update cold-start toast: 90s warning, 30s display duration

Move NeMo and Qwen to GPU on T4: explicit .cuda() for NeMo, device_map=auto + float16 for Qwen

Switch Space hardware to t4-small for persistent GPU inference

Switch from ZeroGPU to persistent CPU Space: remove @spaces.GPU decorators, float32 for Qwen, cpu-basic hardware

Add cold-start toast notification when ASR model loads for first time

Increase ZeroGPU transcription duration to 180s

Revert to lazy loading inside GPU context with duration=120 to avoid Numba JIT timeout

Load models at module level so ZeroGPU context reuses them across calls

Bypass Lhotse path for ZeroGPU inference (use_lhotse=False)

Fix daemonic worker error (num_workers=0) and seed data items_json format

Fix Gradio version to 6.15.2 to match WizardCapture component

Fix wizardcapture install: use sys.path instead of local pip path

Fix NeMo pip install syntax for newer pip

Add HF Space README config

Initial v2: Nemotron ASR + Qwen2.5 parse via Modal (local) and ZeroGPU (Space)