🔍 Multi-Agent Research AI

🚀 Overview

An AI-powered research assistant that answers user queries by combining:

• Retrieval-Augmented Generation (RAG)
• Vector database search (pgvector)
• Real-time web search (Tavily API)
• Multi-agent orchestration using LangGraph

The system retrieves relevant information from both a local knowledge base and the web, then generates a structured response.

---

🧠 How It Works

🔄 Flow

User Query
   ↓
Retrieval Agent (vector search)
   ↓
Router Agent
   ├──→ Synthesis Agent (if context is sufficient)
   └──→ Web Agent → Synthesis Agent
   ↓
Final Answer

🧩 Agents

Retrieval Agent

• Converts the query into embeddings
• Searches the vector database (pgvector)
• Returns relevant stored content

Router Agent

• Checks if retrieved context exists
• Routes execution to the next agent

Web Agent

• Calls Tavily API
• Retrieves relevant web content

Synthesis Agent

• Combines vector and web context
• Generates:
  • Explanation
  • Key insights
  • Final summary

---

🏗️ Tech Stack

Layer	Technology
Backend	FastAPI, LangGraph
Database	PostgreSQL + pgvector
Embeddings	Sentence Transformers (all-MiniLM-L6-v2)
Web Search	Tavily API
Frontend	Streamlit
Infrastructure	Docker, Docker Compose

---

⚙️ Setup

git clone <repo-url>
cd <project-folder>
docker-compose up -d --build

---

🔎 API

POST /research

Request:

{
  "query": "What is RAG?"
}

Response:

{
  "final_answer": "..."
}

---

📂 Project Structure

app/
 ├── main.py
 ├── graph.py
 ├── state.py
 ├── db.py
 ├── vector_store.py
 ├── tools/
 │    ├── embedder.py
 │    └── web_search.py
 └── agents/
      ├── retrieval_agent.py
      ├── web_agent.py
      ├── synthesis_agent.py
      └── router_agent.py

frontend/
 └── streamlit_app.py

---

🧠 Summary

The system processes a query by retrieving relevant context from a vector database, optionally augmenting it with live web results, and generating a structured answer through a multi-agent workflow.