Update README.md

README.md

@@ -9,10 +9,75 @@ sdk_version: 5.33.0
 tags:
 - mcp-server-track
 ---
-## Cursor:
+A powerful web-based tool for comprehensive technical analysis and visualization of AI models. Upload ONNX models or describe your architecture in text, and get deep insights powered by multiple LLMs (Claude, GPT, Gemini, Mistral) with advanced visualizations (SHAP, LIME, Attention, Architecture, Parameters).
+
+## Features
+- **Multi-LLM Analysis:** Choose from Anthropic Claude, OpenAI GPT, Google Gemini, or Mistral for model interpretation.
+- **ONNX & Text Input:** Upload ONNX files or describe your model in plain text/code.
+- **Comprehensive Reports:** Get detailed technical analysis, architecture breakdown, and use-case assessment.
+- **Advanced Visualizations:** Auto-generates SHAP, LIME, attention heatmaps, architecture diagrams, and parameter distributions using Plotly.
+- **Modern Gradio UI:** Intuitive, tabbed interface for easy exploration and comparison.
+
+## Installation
+1. Clone this repository:
+   ```bash
+   git clone https://huggingface.co/spaces/Agents-MCP-Hackathon/Model_Analysis_MCP/
+   cd Model_Analysis_MCP
+   ```
+2. Install dependencies:
+   ```bash
+   pip install -r requirements.txt
+   ```
+
+## Configuration
+This app requires API keys for the LLMs you wish to use. Set the following environment variables (e.g., in a `.env` file):
+- `ANTHROPIC_API_KEY` (for Claude)
+- `OPENAI_API_KEY` (for GPT)
+- `GOOGLE_API_KEY` or `GEMINI_API_KEY` (for Gemini)
+- `MISTRAL_API_KEY` (for Mistral)
+
+Example `.env`:
+```
+OPENAI_API_KEY=sk-...
+ANTHROPIC_API_KEY=sk-ant-...
+GOOGLE_API_KEY=...
+MISTRAL_API_KEY=...
+```
+
+## Usage
+1. Launch the app:
+   ```bash
+   python app.py
+   ```
+2. Open your browser and go to [http://localhost:7860](http://localhost:7860)
+3. Choose to upload an ONNX model or enter a text description of your model.
+4. Select which LLMs and analysis types you want.
+5. Click **Analyze Model** to generate reports and visualizations.
+
+## Example Inputs
+- "ResNet-50 convolutional neural network with 50 layers, batch normalization, and residual connections for image classification"
+- "BERT transformer model with 12 layers, 768 hidden dimensions, and multi-head attention for NLP tasks"
+- "LSTM recurrent neural network with 256 hidden units for time series prediction"
+- "U-Net architecture with encoder-decoder structure and skip connections for image segmentation"
+
+## Demo
+
+### Cursor:
 <video controls autoplay src="https://cdn-uploads.huggingface.co/production/uploads/6615b9363abccce37d27bbcc/EJvYAY7iO0gsD2hA4NRXZ.mp4"></video>
 
-## Gradio UI:
+### Gradio UI:
 <video controls autoplay src="https://cdn-uploads.huggingface.co/production/uploads/6615b9363abccce37d27bbcc/bS0aNHLPR8leU2j87awql.qt"></video>
 
+## Requirements
+See `requirements.txt` for the full list. Key packages:
+- gradio
+- openai, anthropic, mistralai, google-generativeai
+- onnx, onnxruntime
+- numpy, pandas, plotly, networkx
+- python-dotenv
+
+## References
+- [Gradio Documentation](https://gradio.app/docs/)
+- [ONNX](https://onnx.ai/)
+
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference