This project provides a full-stack example of building research-augmented conversational AI agents using Google Gemini 2.5 and LangGraph. It targets developers looking to create sophisticated research tools that can dynamically generate search queries, perform web research, reflect on findings, and synthesize answers with citations. The primary benefit is a working template for complex, multi-step AI workflows.

How It Works

The backend utilizes LangGraph to orchestrate a research agent. This agent iteratively refines its approach by first generating initial search queries with a Gemini model. It then executes these queries via the Google Search API, analyzes the results for knowledge gaps using another Gemini model, and generates follow-up queries if necessary. This reflective, iterative process continues until a comprehensive answer can be synthesized with citations.

Quick Start & Requirements

• Install Dependencies:
  • Backend: cd backend && pip install .
  • Frontend: cd frontend && npm install
• Prerequisites: Node.js, npm (or yarn/pnpm), Python 3.11+, GEMINI_API_KEY.
• Run Development Servers: make dev (runs both frontend and backend).
• Docs: LangGraph Documentation

Highlighted Details

• Fullstack architecture with React (Vite) frontend and LangGraph/FastAPI backend.
• Dynamic search query generation and iterative web research with reflection.
• Generates answers with citations from gathered sources.
• Hot-reloading for both frontend and backend during development.

Maintenance & Community

The project is a Google-maintained quickstart example. Further deployment details and LangGraph specifics can be found in the LangGraph Documentation.

Licensing & Compatibility

• License: Apache License 2.0.
• Compatibility: Suitable for commercial use and integration with closed-source applications.

Limitations & Caveats

Production deployment requires Redis for pub-sub and PostgreSQL for state persistence and task queue management. The provided Docker Compose example also requires a LangSmith API key.