🤖 Multi-Agent Research Assistant

Built by Sai Varshini | GitHub

Reduced a 3-hour manual research workflow to under 2 minutes using a team of 4 AI agents.

📋 Summary

This project is a Multi-Agent Research Assistant that automates the entire research workflow using 4 specialized AI agents. Instead of spending hours manually searching, reading, and writing research reports, this tool does it all in under 2 minutes.

How it works

1. You enter a research topic (and optionally upload a PDF)
2. Research Agent searches the web using Tavily and reads your PDF using RAG
3. Analysis Agent extracts the top 5 key themes and insights
4. Writing Agent produces a structured professional report with all sections
5. Review Agent checks for accuracy, missing points and polishes the final output
6. You get a complete report with citations — downloadable as PDF or Word

---

What makes it special

• Uses RAG (Retrieval Augmented Generation) to process large PDFs intelligently
• Has conversation memory — agents remember your past research
• Saves everything to a local database — research history and PDF library
• Fully containerized with Docker — runs anywhere
• Built with production-grade tools — FastAPI, CrewAI, LangChain
• 100% free to run using Groq free tier and Tavily free tier

---

Built for

• Students who need to research and write reports quickly
• Professionals who need fast summaries of documents and topics
• Developers learning how to build multi-agent AI systems
• Anyone who wants to save hours of manual research time

---

📸 Screenshot

---

✨ Features

• 🔍 4 Specialized AI Agents — Research, Analysis, Writing, and Review agents working in sequence
• 🌐 Real-time Web Search — Tavily API for up-to-date information from the web
• 📄 PDF Ingestion with RAG — Upload PDFs and use vector search to find the most relevant sections
• 🧠 Conversation Memory — Agents remember past research and use it as context
• 📊 Citation Tracking — Sources and references listed in every report
• 💾 Research History — All past reports saved and viewable anytime
• 📚 PDF Library — Track all uploaded PDFs
• ⚙️ Settings Page — Configure LLM model, search results, and PDF limits
• 📥 Export Reports — Download as PDF or Word document
• 🚀 FastAPI Backend — Async endpoints supporting concurrent requests
• 🐳 Docker Ready — Fully containerized for easy deployment
• 🆓 Completely Free — Uses Groq free tier (100,000 tokens/day)

---

🏗 Architecture

User Request
     ↓
FastAPI (port 8000)
     ↓
CrewAI Orchestrator
     ↓
+--------------------------------------------------+
|  Research Agent → Analyst → Writer → Reviewer   |
+--------------------------------------------------+
     ↓                    ↓
Tavily Search          PDF RAG (ChromaDB)
          ↓         ↓
     Groq LLM (Llama 3.3 70B)
               ↓
     Final Report (JSON + UI)

---

🛠 Tech Stack

| Tool | Purpose | |---|---| | CrewAI | Multi-agent orchestration | | LangChain | LLM framework and tooling | | Groq (Llama 3.3 70B) | Fast, free LLM inference | | Tavily API | Real-time web search | | ChromaDB | Vector database for PDF RAG | | Sentence Transformers | PDF chunk embeddings | | FastAPI | Backend REST API | | Uvicorn | ASGI server | | Docker | Containerization | | ReportLab | PDF export | | python-docx | Word export | | Jinja2 | HTML templating |

---

🚀 Quick Start

Prerequisites

• Python 3.11
• Free Groq API key
• Free Tavily API key

Installation

1. Clone the repository:

git clone https://github.com/Saivarshini-001/research-assistant.git
cd research-assistant

2. Create and activate virtual environment:

python3.11 -m venv venv
source venv/bin/activate

3. Install dependencies:

pip install -r requirements.txt

4. Configure API keys:

Open config.py and add your keys:

TAVILY_API_KEY = "your-tavily-api-key"
GROQ_API_KEY = "your-groq-api-key"

5. Run the application:

uvicorn main:app --reload

6. Open in browser: http://127.0.0.1:8000/ui

🐳 Docker Deployment

Build the image:

docker build -t research-assistant .

Run the container:

docker run -p 8000:8000 research-assistant

Open in browser: http://127.0.0.1:8000/ui

📁 Project Structure

research-assistant/
  main.py            - FastAPI server + endpoints
  agents.py          - 4 CrewAI agent definitions
  tasks.py           - Task definitions for each agent
  tools.py           - Tavily search + PDF reader tools
  config.py          - API keys + LLM configuration
  export.py          - PDF + Word export generation
  database.py        - Local JSON database for history
  rag.py             - RAG system for PDF processing
  requirements.txt   - Python dependencies
  Dockerfile         - Docker configuration
  .gitignore         - Git ignore rules
  templates/
    index.html       - Full web UI

---

🔌 API Endpoints

| Method | Endpoint | Description | |---|---|---| | GET | /ui | Web interface | | GET | / | Health check | | POST | /research | Run research query | | GET | /history | Get all past reports | | GET | /history/{id} | Get specific report | | GET | /pdfs | Get PDF library | | GET | /settings | Get settings | | POST | /settings | Update settings | | GET | /export/pdf | Export last report as PDF | | GET | /export/docx | Export last report as Word |

---

💡 Usage Examples

Research a topic:

curl -X POST http://localhost:8000/research \
  -F "query=What is the impact of AI on healthcare?"

Research with PDF:

curl -X POST http://localhost:8000/research \
  -F "query=Summarize this document" \
  -F "pdf=@your_document.pdf"

---

⚡ Performance

• Reduced 3-hour manual research workflow to under 2 minutes
• Tested across 20+ research prompts
• Supports concurrent requests via async FastAPI endpoints
• Average response time: 15-30 seconds with Groq
• PDF RAG processes documents of any size using vector search

---

⚠️ Limitations

• Free tier token limits — Groq allows 100,000 tokens/day. Create multiple free accounts to rotate keys.
• Response time — Reports take 15-30 seconds with Groq cloud. Local Ollama is slower (10-20 mins).
• Sources accuracy — Agents use Tavily web search which may not always find the most authoritative sources.
• Internet required — Tavily search needs internet. Offline mode uses only PDF content.

---

🔧 Configuration

You can switch LLM providers by changing the model in agents.py:

# Groq (fast, free)
llm="groq/llama-3.3-70b-versatile"

# Google Gemini (more tokens)
llm="gemini/gemini-2.0-flash"

# Local Ollama (completely offline)
llm="ollama/mistral"

---

🙏 Acknowledgements

• CrewAI — amazing multi-agent framework
• Groq — incredibly fast free LLM inference
• Tavily — best search API for AI agents
• FastAPI — modern Python web framework
• ChromaDB — simple local vector database

---

📝 License

MIT License — feel free to use this project for learning and building!