MATBot: MATLAB AI Troubleshooting Assistant

Live Demo

👉 Try MATBot Now

Overview

MATBot is an intelligent assistant designed to troubleshoot MATLAB software-related issues. The application leverages RAG (Retrieval-Augmented Generation) architecture and LLM models to provide accurate, contextual solutions to common MATLAB problems by drawing from official MathWorks documentation.

Features

• AI-Powered Troubleshooting: Identifies and resolves MATLAB errors with step-by-step solutions
• Semantic Clustering: Organizes technical documentation into relevant clusters for better retrieval
• Interactive Chat Interface: User-friendly Streamlit-based chat interface with real-time feedback
• Model Customization: Select between different Gemini models with configurable system prompts
• Response Evaluation: Automatic quality assessment of AI responses with detailed feedback
• Structured Content Output: Presents solutions in an easy-to-follow format with problem summary, root cause, and resolution
• Self-Memory System: Learns from previous high-quality interactions to provide better responses
• Adaptive Response Style: Automatically detects whether to provide concise or detailed answers
• Visual Support: Provides relevant images to help understand complex MATLAB concepts
• Query Enhancement: Uses HyDE (Hypothetical Document Embeddings) to improve search quality

Architecture

The application follows a modular architecture:

1. Data Collection & Processing:

   • Web scraping of MATLAB documentation
   • HTML to structured JSON conversion (structurify.py)

2. Knowledge Base:

   • Semantic clustering of documents (clustering.py)
   • FAISS vector index for efficient similarity search

3. RAG Engine:

   • Query processing
   • Relevant document retrieval
   • Context-enhanced generation

4. Agent System:

   • Debugger agent for problem-solving (debugger_agent.py)
   • Evaluator agent for quality assessment (evaluator_agent.py)
   • Intent agent for response type detection (intent_agent.py)
   • Concise agent for summarization (concise_agent.py)

5. User Interface:

   • Streamlit web application (streamlit_chat_app.py)
   • Interactive RAG configuration parameters
   • Admin dashboard for system monitoring

Technical Stack

• Python 3.10+
• Machine Learning: scikit-learn, sentence-transformers, FAISS
• LLM APIs: Google Gemini
• Web Framework: Streamlit
• Data Processing: BeautifulSoup, pandas
• Vector Database: FAISS (Facebook AI Similarity Search)
• Database: PostgreSQL (via Neon DB)

Getting Started

Prerequisites

• Python 3.10+
• API key for Google Gemini

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/matbot.git
   cd matbot

2. Install dependencies:

   pip install -r requirements.txt

3. Set up environment variables:

   • Create a .env file in the project root
   • Add your API key:

     GEMINI_API_KEY=your_gemini_api_key
     

Running the Application

Launch the chat interface:

streamlit run Landing.py

Usage

1. Enter your MATLAB-related troubleshooting query in the chat input
2. The system will:
   • Retrieve relevant documentation from the knowledge base
   • Generate a structured response with problem analysis and solution
   • Evaluate the quality of the response
3. Use the model parameters sidebar to configure the LLM and RAG parameters before starting the chat

Project Structure

• /agents/: LLM agent implementations
• /utils/: Utility functions and wrappers for LLM APIs
• /corpus/: Processed knowledge base (generated)
• /pages/: Streamlit pages for the application
• structurify.py: HTML processing utilities
• clustering.py: Document clustering implementation
• Landing.py: Main entry point and authentication
• requirements.txt: Project dependencies

Future Improvements

• Expanded knowledge base with additional MATLAB documentation
• Integration with MATLAB for direct code analysis
• User feedback-based continuous learning
• Support for more LLM providers
• Enhanced visualization of technical solutions
• Mobile-friendly responsive design
• Fine-tuned domain-specific embeddings

License

This project was created as part of the HCL Tech CS671 Hackathon.