Project X is an advanced research assistant designed to streamline and enhance the research process by leveraging state-of-the-art AI technologies. With hybrid retrieval, multi-modal support, and advanced citation capabilities, Project X empowers researchers to efficiently analyze, retrieve, and cite information from diverse data sources.
- Hybrid Retrieval: Combines full-text and vector-based retrieval with re-ranking to ensure the highest quality of results.
- Multi-File Search: Search data across multiple documents, not limited to a single file.
- AI Agent Assistance: If local data is insufficient, a mode with AI agents can perform searches on the web.
- Multi-Agent Architecture: Powered by Autogen and Mistral AI, enabling dynamic and context-aware retrieval-augmented generation (RAG). The agents are designed to collaborate seamlessly by performing advanced tasks:
- Web Search: Agents can autonomously search the web to gather real-time, relevant information for user queries.
- Research Paper Retrieval: By querying Arxiv databases and repositories, agents can locate and summarize research papers or articles pertinent to the topic of interest.
- Knowledge Aggregation: Through intelligent filtering and integration of diverse data sources, the agents ensure high-quality and accurate responses tailored to the user's needs.
- Document Parsing: Perform QA on documents containing text, figures, and tables.
- Multi-Document Support: Seamlessly analyze and extract insights from multiple documents simultaneously.
- Detailed Citations with Previews:
- Ensure correctness with detailed citations directly in the in-browser PDF viewer.
- View relevant scores and highlights for context.
- Receive warnings if retrieval pipelines return low-relevance articles.
- Video Citations (Future Feature):
- Cite specific quotes from videos with accurate timestamps and proper format.
- Save time on manual video referencing, ensuring precision in both time and content.
- AI-Powered Generation: Create mind maps automatically from any topic or text using Mistral AI.
- Interactive Editing:
- Expand nodes to explore related concepts
- Delete nodes and their connections
- Visual node selection and highlighting
- Dynamic Visualization: Interactive graph layout with physics simulation for optimal readability.
- Automatically generate visualizations from data files using Python and matplotlib.
- Interactive plots and charts to support research findings and presentations.
- Streamlit: Modern Python web framework for interactive UI components
- Streamlit-Agraph: Interactive graph visualization for mind maps
- Custom CSS: Responsive design with modern animations and styling
- Mistral AI: Large language model for intelligent responses and mind map generation
- Snowflake Cortex Search Service: Enterprise data platform for efficient RAG implementation including pre-processing, labelling, storing documents, and searching relevant text chunks.
- Trulens: Evaluation and monitoring framework for AI applications
- Python 3.8+: Core programming language
- User Proxy: Initial query handling
- Intent Classifier: A semantic router to identify user's intent, whether reading uploaded documents, searching for papers, searching for real-time data, or general query.
- Specialized Agents:
- Document Reading Agent: extracts and retrieves relevant information from uploaded documents using LAYOUT mode, embeddings, and search algorithms.
- Web Search Agent: fetches real-time data via search engine APIs or scraping, with result ranking for relevance.
- Articles Research Agent: searches academic databases for papers relevant to the query.
- Writer Agent: aggregates context from all agents and generates a cohesive, factually accurate response. Prioritizes data based on agent reliability and relevance.
- Critic Agent: Response refinement based on user's query, retrieved context information, writer agent's response to reduce hallucinations from LLMs.
Utilizing Trulens for tracking and evaluating performance across three core metrics:
- Answer Relevance: Ensures responses are coherent, accurate, and reasoned.
- Context Relevance: Validates the quality of retrieved information, ensuring it aligns with the query.
- Groundedness: Functions as hallucination detection, verifying all responses against source documents.
Our development process included creating multiple RAG versions:
- Agent-Based Version: Combines multiple AI agents (Document Reading, Article Research, Web Search) with Trulens' context filter guardrails. This approach enhanced context quality, reduced hallucinations, and balanced latency.
- Agent-Free Version: Offers simplicity with low latency by retrieving relevant context solely from uploaded documents. Although limited in accessing real-time data, it maintains high context relevance.
Our agent_v3's iterative refinements, including Trulens context filtering at threshold 0.5 and enhanced processing through multiple AI agents, make it a more robust and efficient RAG system compared to Agent_v1. It achieves better groundedness and higher answer quality.
- Python 3.8 or higher
- Required Python libraries (see
requirements.txt) - Mistral AI API key, OPENAI API key
- Snowflake account configuration
- Clone the repository:
git clone https://github.com/stephanienguyen2020/project-x.git
cd project-x- Create and activate a virtual environment:
python -m venv .venv
source .venv/bin/activate # On Windows use: venv\Scripts\activate- Install dependencies:
pip install -r requirements.txt-
Set up your secrets: Create .streamlit/secrets.toml following the structure in secrets.toml.sample
-
Run the application:
streamlit run app.pyThe application will be available at http://localhost:8501
project-x/
├── assistance # Multi-agents
│ ├── ...
│ ├── writer_agent.py # Main researcher agent
│
├── components/ # UI components
│ ├── chatbot.py # Chat interface
│ ├── info_panel.py # Information panel
│ ├── ...
│ ├── mindmap.py # Mind map visualization
│ └── settings.py # Settings interface
│
├── services/ # Backend services
│ └── rag_agents.py # RAG version with agents
│ └── rag_no_agents.py # RAG Version without agents
├── ...
├── utils/ # Utility functions
│ └── ...
│ └── trulens_feedbacks.py # Trulensfeedback metrics
│ └── trulens_utils.py # Trulens utilities
│
├── prompts/ # System prompts
├── app.py # Main Streamlit application
└── requirements.txt # Dependencies
-
Connection Issues
- Verify your Mistral AI API key is correct
- Check Snowflake credentials and network access
- Ensure all required environment variables are set
-
Visualization Problems
- Make sure matplotlib and streamlit-agraph are properly installed
- Check browser console for any JavaScript errors
- Try clearing browser cache and Streamlit cache
-
Performance Issues
- Adjust Snowflake warehouse size if needed
- Consider reducing chunk size for large documents
- Monitor memory usage for large mind maps
- Fork the repository
- Create a feature branch (
git checkout -b feature/amazing-feature) - Commit your changes (
git commit -m 'Add amazing feature') - Push to the branch (
git push origin feature/amazing-feature) - Open a Pull Request
This project is licensed under the MIT License - see the LICENSE file for details.
For support, please:
- Check the Issues page
- Review existing documentation
- Create a new issue if needed
- Mistral AI for their powerful language model
- Snowflake for enterprise data platform capabilities
- Streamlit team for the excellent web framework
- All contributors who have helped shape this project