An Assembly Line of AI Agents for Company Due Diligence

What it does Built on LangGraph, this web application automates deep-dive company research by orchestrating a pipeline of specialized agents. You submit a company name, and the agents—each assigned a slice of the investigation—gather, filter, and synthesize data into a comprehensive markdown report. A React frontend tracks progress and serves the final output, which can also be exported as a PDF.

The interesting bit Instead of trusting one model to do everything, the architecture splits duties by strength: Gemini 2.5 Flash handles the high-context synthesis of raw research into category briefings, while GPT-5.1 takes over as the persnickety editor for final formatting, deduplication, and markdown structure. It is essentially an assembly line where each station is a different LLM with a different job description.

Key highlights

• LangGraph-powered pipeline with four research agents (CompanyAnalyzer, IndustryAnalyzer, FinancialAnalyst, NewsScanner) feeding a collector and curator
• Tavily search provides relevance scoring for content filtering, with a configurable threshold to weed out low-quality sources
• Dual-model backend: Gemini 2.5 Flash for briefing generation, GPT-5.1 for report editing and real-time streaming
• FastAPI backend with async polling; React frontend shows live progress and supports PDF export
• Live demo already running at companyresearcher.tavily.com

Caveats

• The backend relies on a simple polling mechanism rather than true streaming or push updates for progress tracking
• You will need four separate API keys (Tavily, Gemini, OpenAI, Google Maps) plus both Node.js and Python environments to self-host
• MongoDB persistence is optional, so out-of-the-box state storage depends on your setup

Verdict A solid scaffold for developers who need structured, multi-source company intelligence without building an agent framework from scratch. Not for anyone trying to avoid third-party API bills or complex environment wrangling.