Multi-Agent Research System - Autonomous Web Researcher

AI Platform (FastAPI + Next.js + LangGraph)

About

Enterprise-grade AI research assistant that orchestrates 4 specialized agents (Planner, Researcher, Analyst, Writer) to autonomously conduct web research, synthesize findings, and generate comprehensive markdown reports with real-time WebSocket streaming.

Outcomes & Impact

My Role

Built the complete full-stack platform: FastAPI backend with LangGraph agent orchestration, Next.js 14 frontend with real-time streaming UI, and 4 specialized AI agents with conditional looping.

Core Capabilities

• 4 specialized agents: Planner (task decomposition), Researcher (web search + scraping), Analyst (synthesis + gap detection), Writer (report generation)
• Real-time WebSocket streaming of agent steps, tool calls, and findings to the UI
• PDF/DOCX document upload and in-memory search during research
• MCP (Model Context Protocol) tool integration for extensibility
• Conditional graph loops: Analyst can request additional research if gaps detected
• Multi-format report generation with proper citations

Architecture

• LangGraph state machine with conditional edges and max 2 research loops
• FastAPI async backend with thread-safe queue for WebSocket event streaming
• Next.js 14 frontend with TypeScript, Tailwind CSS, react-markdown
• Tavily API for deep web search, BeautifulSoup4 for URL scraping
• PyMuPDF + python-docx for document processing

Tech Stack