Research Assistant

Overview

A multi-agent system that performs automated research and generates comprehensive responses. A research agent searches and analyzes sources, an answer drafter turns those findings into a cited response, and a LangGraph workflow coordinates the two.

Features

• Research agent that gathers information using the Tavily search API
• Search results analyzed with the Gemini model, returning structured research findings
• Answer drafter agent that takes the research findings and synthesizes them into clear responses
• Citations to sources included in every drafted response
• Responses can be refined based on feedback
• LangGraph workflow orchestration coordinating the research and drafting agents
• Workflow state transitions maintained across the graph
• The graph is compiled and executed asynchronously
• Pydantic models validate the workflow state
• WorkflowState tracks the topic, research results, draft results and next state
• CLI interface — enter a research topic when prompted

Architecture

• Research Agent — uses the Tavily search API to gather information, analyzes the search results using the Gemini model, and returns structured research findings.
• Answer Drafter Agent — takes the research findings as input, synthesizes them into clear responses, includes citations to sources, and can refine responses based on feedback.
• Workflow Orchestration — uses LangGraph for state management, coordinates the research and drafting agents, maintains workflow state transitions, and compiles and executes the graph asynchronously.

Implementation details

• State management — Pydantic models validate state; WorkflowState tracks the topic, research results, draft results and next state.
• Graph flow — the entry point is the "research" node, where the research agent performs search and analysis; results pass to the "draft" node, where the answer drafter generates the final response, and the graph completes with the final state.
• Async execution — uses LangGraph's ainvoke pattern with a proper async/await flow throughout, with state transitions handled by graph compilation.

Key files

• main.py — entry point and CLI interface
• graph/workflow.py — LangGraph workflow implementation
• agents/research_agent.py — research agent implementation
• agents/answer_drafter.py — answer drafting agent implementation
• utils/tavily_tool.py — search API integration

Configuration

• GOOGLE_API_KEY — for Gemini model access
• TAVILY_API_KEY — for search functionality

Setup & usage

• Install dependencies: pip install -r requirements.txt
• Set up the environment variables in a .env file.
• Run the assistant: python main.py
• Enter your research topic when prompted.
• The system performs research using Tavily, analyzes the findings with Gemini, generates a comprehensive response, and provides source citations.

Model configuration

Uses the Gemini 1.5 Flash model for research analysis, response drafting, and content refinement.