Summary

This project offers an enhanced fullstack application for deep research, using LangGraph and Google Gemini to automate comprehensive web research. It targets developers and researchers building advanced conversational AI agents capable of dynamic query generation, iterative web searching, reflective analysis, and citation-backed answer synthesis. The benefit is an optimized agent workflow and frontend display, streamlining information gathering and presentation.

How It Works

The backend uses a LangGraph agent orchestrated via FastAPI. The agent generates initial search queries with Gemini, performs web research using the Google Search API, and employs Gemini for reflective analysis to identify knowledge gaps. This process iteratively refines queries and gathers information up to a configured maximum. Gemini then synthesizes the collected data into a coherent, cited answer. This enables robust, multi-step reasoning and information retrieval.

Quick Start & Requirements

• Primary Install/Run: Use make dev from the project root for simultaneous backend/frontend development server startup.
• Prerequisites: Node.js, npm (or yarn/pnpm), Python 3.8+. A Google Gemini API key (GEMINI_API_KEY) is mandatory for the backend. For Docker deployment, a LangSmith API key (LANGSMITH_API_KEY) is also needed.
• Dependencies: Install backend Python dependencies via pip install . in backend/ and frontend Node.js dependencies via npm install in frontend/.
• Links: Technical documentation is in docs/document-generation-flow.md (English) and docs/document-generation-flow-ZH.md (Chinese). Contact details and blog are provided.

Highlighted Details

• Features a fullstack architecture with React/Vite frontend and LangGraph/FastAPI backend.
• Employs Google Gemini for intelligent query generation, web research, reflection, and answer synthesis.
• Implements dynamic search query generation and reflective reasoning to identify and address knowledge gaps.
• Provides answers with verifiable citations sourced from web research.
• Includes optimizations for agent workflow and frontend display effects.

Maintenance & Community

Authored by Peng.G, contactable via 𝕏 (@Stephen4171127), email (foreveryh@gmail.com), or WeChat (browncony999). A personal blog is at https://me.deeptoai.com. No specific community channels (Discord/Slack) are mentioned.

Licensing & Compatibility

Licensed under the Apache License 2.0. This permissive license generally allows commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

The README does not explicitly detail limitations or known issues. The agent's research depth is implicitly bounded by a configured maximum number of iterative refinement loops. The project is presented as a demonstration and example, suggesting potential need for further hardening for production environments.