LangGraph Researcher

This project implements a dual-agent system for deep research using Tavily, LangChain and LangGraph. It features a Research Agent for web crawling and data collection, and an Answer Drafter Agent for synthesizing answers from gathered information.

Requirements

• Python 3.10 or higher
• Ollama installed and running locally (for LLMs)
• A free Tavily API key

Tech Stack & How Each Component Is Used

• LangChain: Provides the agent/tool abstraction and LLM orchestration. Used for integrating the LLM (Ollama) and custom tools (Tavily search).
• LangGraph: Manages the agent workflow as a directed graph. Each node is a functional agent (e.g., research, drafting). Edges define the sequence (research → draft answer → END). See main_graph.py for the workflow implementation.
• Tavily: Supplies live, up-to-date web data for research queries. Wrapped as a LangChain tool in tools/tavily_tool.py.
• Ollama: Local LLM backend (phi3, llama3, mistral, etc.) for answer synthesis. Used in agents/answer_drafter_agent.py.
• Streamlit: Provides a modern, interactive web UI (app.py).

How It Works (System Flow)

1. User submits a query (via CLI or web UI).
2. Research Agent (see agents/research_agent.py) uses Tavily to crawl the web and structure findings.
3. Answer Drafter Agent (see agents/answer_drafter_agent.py) takes the structured research and synthesizes a detailed answer using Ollama LLM.
4. LangGraph (main_graph.py) orchestrates the workflow: research node → drafting node → END.
5. Results are displayed in the CLI or web UI.

Overview

This system is a dual-agent AI research assistant, built with LangChain, LangGraph and Tavily. It automates deep online research and answer drafting using two collaborating agents:

• Research Agent: Gathers and organizes web information using Tavily (web crawling/search API).
• Answer Drafter Agent: Synthesizes a comprehensive answer from the research results.

Architecture

User Query
   |
   v
[Research Agent] --(Tavily Search)--> [Research Results]
   |
   v
[Answer Drafter Agent] --(LLM Synthesis)--> [Drafted Answer]

• LangChain is used for agent/tool integration and LLM orchestration.
• LangGraph organizes the research and drafting process as a graph workflow.
• Tavily provides live web data for research.

Features

• Research Agent: Gathers and organizes online information using Tavily API
• Answer Drafter Agent: Synthesizes and drafts answers using a local Ollama LLM (phi3, llama3, or mistral)
• LangChain & LangGraph orchestration
• Modular agent design: easy to add more agents (e.g., fact-checker, summarizer)
• Extensible workflow via LangGraph
• Clean, modern web UI (Streamlit) for demonstration

Setup

1. Clone this repo
2. Install dependencies:

   pip install -r requirements.txt

3. Set up your Tavily API key in a .env file:

   TAVILY_API_KEY=your_tavily_api_key

Project Structure

• agents/ - Agent implementations
• tools/ - Custom tools (Tavily wrapper, etc.)
• main.py - CLI entry point
• main_graph.py - LangGraph workflow
• app.py - Streamlit web app
• requirements.txt - Dependencies

Web UI (Streamlit)

A modern Streamlit app (app.py) allows users to interact with the system via a clean web interface. Enter a query, see research results and get a drafted answer, all in one place.

Troubleshooting

• Make sure Ollama is running (ollama run phi3).
• Ensure your .env file contains a valid Tavily API key.
• If you see parsing errors, check that you are not using an old agent setup.
• For dependency issues, run pip install -r requirements.txt again.

Example Use Case

1. User enters: "What are the latest breakthroughs in quantum computing?"
2. Research Agent crawls the web for up-to-date info.
3. Answer Drafter Agent synthesizes a detailed, referenced answer.
4. User sees both the sources and the drafted summary in the UI.

How to Run

• CLI: python main_graph.py
• Web: streamlit run app.py

Extensibility

• Add more agents (e.g., fact-checker, summarizer) by extending the graph.
• Swap LLMs or add retrieval-augmented generation.
• Add a web or chat UI.
• To use a different Ollama model, change the model name in agents/answer_drafter_agent.py.

Extending the System

• Add new agents (e.g., fact-checker, summarizer) by implementing new classes and updating the workflow graph.
• Enhance research result structuring for better answer quality.
• Integrate with other LLMs or APIs as needed.