How pogjester/company-research-agent works
Project Type and Purpose
• This is a full-stack web application—a multi-agent AI research platform that gathers, filters, and compiles company research into a structured report.
• Frontend: React (Vite + TypeScript + Tailwind) for real-time UI and progress streaming.
• Backend: Python FastAPI orchestrating a pipeline of specialized “nodes” (researchers, collector, curator, briefing, editor).
• External services: Tavily AI search, Google Gemini API, OpenAI API, (optional) MongoDB persistence.
File Structure
• Top-level:
• application.py / langgraph_entry.py — FastAPI entrypoints.
• backend/ — business logic:
– classes/ (state, graph orchestration)
– nodes/ (researcher nodes, processing nodes: collector, curator, briefing, editor)
– services/ (mongodb connector, PDF generator, WebSocket manager)
– utils/ (reference handling, general helpers)
• ui/ — React app: src/components, styles, utils, Vite config, .env examples.
• docker-compose.yml, Dockerfile — containerized deployment for both backend and frontend.
• Key config: requirements.txt, package-lock.json, docker-compose.yml.
README Insights
• Agentic pipeline: CompanyAnalyzer, IndustryAnalyzer, FinancialAnalyst, NewsScanner → Collector → Curator → Briefing (Gemini 2.0 Flash) → Editor (GPT-4.1-mini).
• Real-time streaming via WebSocketManager in backend/services.
• Dual-model approach for context-heavy vs precision formatting.
• Optional MongoDB for job persistence.