Update README.md

README.md

@@ -1,12 +1,96 @@
 ---
 title: Voice OCR Agent
 emoji: 👁
-colorFrom: green
-colorTo: gray
+colorFrom: purple
+colorTo: indigo
 sdk: gradio
 sdk_version: 5.45.0
 app_file: app.py
 pinned: false
+license: mit
+tags:
+  - speech-recognition
+  - whisper
+  - zero-shot
+  - ocr
+  - summarization
+  - question-answering
+  - ai-agent
+  - gradio
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# 🎤🧾 Multimodal Voice & OCR Agent
+**Voice commands → intents → tools** and **images → text → summary → QA** — using only **pre-trained models** on Hugging Face.
+
+> **Live demo:** open the **App** tab above.  
+> Works on CPU (tiny models) and GPU (faster, larger models).
+
+---
+
+## 🔎 Overview
+This Space demonstrates a simple **agent loop** across two modalities:
+
+- **Voice Agent tab:** microphone/upload → ASR → **zero-shot** intent detection → run a mapped tool → append to an execution log.
+- **OCR tab:** image/PDF page → OCR → summarization → optional **question answering** over the extracted text.
+
+---
+
+## 🧩 Models Used (pre-trained)
+- **ASR (speech→text):** `openai/whisper-tiny` *(use `openai/whisper-small` on GPU)*  
+- **Zero-shot intent:** `facebook/bart-large-mnli` *(multilingual alt: `MoritzLaurer/mDeBERTa-v3-base-xnli-multilingual-nli-2mil7`)*  
+- **OCR:** `microsoft/trocr-small-printed` *(for handwriting: `microsoft/trocr-small-handwritten`)*  
+- **Summarization:** `sshleifer/distilbart-cnn-12-6`  
+- **Question Answering:** `deepset/roberta-base-squad2`  
+
+All models are loaded with `transformers.pipeline` — no training required.
+
+---
+
+## 🧭 How It Works
+1. **Capture:** Gradio handles microphone/file input (audio or image).  
+2. **Perceive:**  
+   - Audio → Whisper ASR → transcript  
+   - Image → TrOCR → extracted text  
+3. **Understand:**  
+   - Transcript → zero-shot classifier over user-editable intents  
+   - OCR text → optional summarizer + QA  
+4. **Act:** Chosen intent maps to a simple tool (e.g., `turn_on_lights`, `set_timer`) and logs the result.
+
+---
+
+## 🧪 Try It
+- **Voice:** say “turn on the lights / set a timer / pause the music”.  
+- **OCR:** upload a screenshot/document → see extracted text + summary, then ask “What’s the due date?” etc.
+
+---
+
+## 🔧 Configuration — Swap Pre-Trained Models
+Change model IDs in `app.py` **or** set Space **Variables** (Settings → Variables) without code changes:
+
+| Component | Env var | Default | Common alternatives |
+|---|---|---|---|
+| ASR | `ASR_MODEL` | `openai/whisper-tiny` | `openai/whisper-small` (GPU) |
+| Zero-shot Intent | `ZSC_MODEL` | `facebook/bart-large-mnli` | `MoritzLaurer/mDeBERTa-v3-base-xnli-multilingual-nli-2mil7` |
+| OCR | `OCR_MODEL` | `microsoft/trocr-small-printed` | `microsoft/trocr-small-handwritten` |
+| Summarizer | `SUM_MODEL` | `sshleifer/distilbart-cnn-12-6` | `facebook/bart-large-cnn` (GPU) |
+| QA | `QA_MODEL` | `deepset/roberta-base-squad2` | any SQuAD2-style model |
+
+**Example (env var):** set `ASR_MODEL=openai/whisper-small` to speed up on GPU.
+
+---
+
+## ⚙️ Requirements
+This Space installs from `requirements.txt`:
+```txt
+transformers>=4.41.0
+torch
+torchaudio
+gradio>=4.0.0
+librosa
+soundfile
+Pillow
+---
+
+System packages in apt.txt:
+
+ffmpeg