A comprehensive research assistant that combines RAG (Retrieval-Augmented Generation) with web search to provide well-sourced, in-depth research reports.

• Research Planning: Automatically breaks down complex questions into focused sub-questions
• Multi-Query Search: Generates multiple search query variations for comprehensive coverage
• Structured Reports: Produces well-organized research reports with:
  • Executive Summary
  • Key Findings
  • Detailed Analysis
  • Conclusion

• Upload PDF or text files to build a knowledge base
• Semantic search across uploaded documents
• Automatic document chunking and vector storage

• Inline Citations: Every claim is backed by source citations [1], [2], or [Web Search]
• Source Tracking: Only shows sources that were actually cited in the response
• Transparent Attribution: Clear distinction between uploaded documents and web search results

• Automatic web search when uploaded documents don't contain answers
• Multi-query strategy for diverse information gathering
• Seamless integration with RAG results

• LangChain: Agent framework and RAG implementation
• LangGraph: Multi-step research workflow
• ChromaDB: Vector store for document embeddings
• Streamlit: Web interface
• OpenRouter: LLM API access (supports multiple models)
• DuckDuckGo: Web search capabilities

1. Install dependencies:

cd backend
pip install -r requirements.txt

2. Set up environment variables in .env:

OPENROUTER_API_KEY=your_api_key_here

3. Run the application:

streamlit run backend/app.py

1. Upload Documents (optional): Add PDF or text files to build your knowledge base
2. Select Model: Choose your preferred model provider (Gemini or OpenRouter)
3. Ask Questions: The agent will:
   • Plan the research approach
   • Search your documents and the web
   • Generate a comprehensive research report with citations

The agent generates structured research reports with proper citations:

# Research Report

## Executive Summary
[Overview with citations]

## Key Findings
[Main points with inline citations [1], [2], [Web Search]]

## Detailed Analysis
[In-depth exploration with sources]

## Conclusion
[Summary]

## Sources
[1] File: document.pdf (page 1)
[Web Search] Web Search

1. Planning: Breaks down complex questions into sub-questions
2. Query Generation: Creates multiple search query variations
3. Retrieval: Searches uploaded documents (RAG) and the web
4. Synthesis: Combines information into a structured report
5. Citation: Tracks and displays only sources actually used