GraphMind - Agentic TensorFlow Debugger

AI-Powered TensorFlow Debugging with ReAct Reasoning & Self-Learning capabilities.

Features

• ReAct Reasoning Framework 🧠 - Advanced reasoning with Thought-Action-Observation cycles
• Knowledge Graph Integration 🕸️ - Entity relationship mapping for better context
• Self-Reflection 🔄 - Quality assessment and confidence scoring
• Continuous Learning 📈 - System improves from user feedback
• Agentic Updates 🤖 - Intelligent knowledge base expansion

Installation

Prerequisites

• Python 3.8 or higher
• pip package manager

Setup

1. Clone or navigate to the project directory:

cd /Users/gowtham/CascadeProjects/GraphMind/GraphMind

2. Create a virtual environment (recommended):

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

4. Ensure artifacts are present:

Make sure the following files exist in the project directory:

• embeddings.npy
• faiss_index.index
• processed_docs.json
• kg_networkx.gpickle

Usage

Starting the Server

python server.py

The server will start on http://localhost:5000

Using the Web Interface

1. Open your browser and navigate to http://localhost:5000
2. Enter your TensorFlow error message in the input field
3. Click "Analyze Error" to get AI-powered debugging assistance
4. Review the solution with:
   • Root cause analysis
   • Step-by-step fixes
   • Code examples
   • Confidence scores
5. Provide feedback to help the system learn

API Endpoints

Analyze Error

POST /api/analyze
Content-Type: application/json

{
  "error_message": "Your TensorFlow error here",
  "session_id": "optional-session-id"
}

Submit Feedback

POST /api/feedback
Content-Type: application/json

{
  "session_id": "session-id-from-analyze",
  "worked": true,
  "new_error": "Optional additional info if it didn't work"
}

Get Learning Statistics

GET /api/learning/stats

Apply Learning Updates

POST /api/learning/apply

Get System Statistics

GET /api/system/stats

Configuration

API Key

Update the OPENROUTER_API_KEY in app_core.py with your OpenRouter API key:

OPENROUTER_API_KEY = "your-api-key-here"

Get your free API key from OpenRouter

Model Selection

Change the model in app_core.py:

MODEL_NAME = "nvidia/nemotron-nano-9b-v2:free"  # or any other OpenRouter model

Artifacts Directory

If your artifacts are in a different location, update:

ARTIFACTS_DIR = "./path/to/artifacts"

Architecture

Core Components

1. OpenRouterLLM - API client for language model interactions
2. ReasoningPlanner - Plans reasoning strategy using ReAct framework
3. EnhancedRetriever - Multi-stage retrieval with knowledge graph awareness
4. AdvancedReActAgent - Main agent with reasoning and reflection
5. FeedbackLearningSystem - Handles user feedback and system updates

Files Structure

GraphMind/
├── app_core.py          # Core logic (no UI dependencies)
├── server.py            # Flask server with API endpoints
├── templates/
│   └── index.html       # Web interface
├── static/
│   ├── css/
│   │   └── style.css    # Styling
│   └── js/
│       └── app.js       # Frontend JavaScript
├── requirements.txt     # Python dependencies
├── README.md           # This file
├── embeddings.npy      # Document embeddings
├── faiss_index.index   # FAISS vector index
├── processed_docs.json # Knowledge base documents
└── kg_networkx.gpickle # Knowledge graph

Learning System

The system continuously improves through:

1. Feedback Collection - Users report if solutions worked
2. ReAct Analysis - AI analyzes failures to extract insights
3. Knowledge Base Updates - New documents added from validated feedback
4. Graph Enhancement - Entity relationships strengthened
5. Embedding Updates - FAISS index expanded with new knowledge

Troubleshooting

Artifacts Not Found

Ensure all artifact files are in the correct directory:

ls -la *.npy *.index *.json *.gpickle

Port Already in Use

Change the port in server.py:

app.run(debug=True, host='0.0.0.0', port=5001)  # Use different port

API Errors

Check your OpenRouter API key is valid and has credits.

Memory Issues

For large knowledge bases, increase Python memory:

export PYTHONMALLOC=malloc

Development

Running in Debug Mode

The server runs in debug mode by default. Disable for production:

app.run(debug=False, host='0.0.0.0', port=5000)

Adding New Features

1. Core logic goes in app_core.py
2. API endpoints go in server.py
3. UI updates go in templates/index.html and static/

License

MIT License

Credits

• Powered by NVIDIA Nemotron via OpenRouter
• Built with Flask, FAISS, and NetworkX
• Uses Sentence Transformers for embeddings