ResearchMind — Full-stack AI Research Assistant

React + TypeScript + Vite frontend, FastAPI backend, LangChain, OpenAI, and Tavily.

Architecture

• frontend/: research dashboard and typed API client
• backend/app/main.py: FastAPI application and REST/SSE endpoints
• backend/app/services/pipeline.py: async job orchestration and progress events
• backend/app/agents/agents.py: search/reader agents and writer/critic chains
• backend/app/tools/web_tools.py: Tavily search and safe URL scraping
• backend/app/core/config.py: environment configuration

Requirements

• Python 3.11+
• Node.js 20+
• OpenAI API key
• Tavily API key

Backend setup (PowerShell)

cd backend
py -3.11 -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install -r requirements.txt
Copy-Item .env.example .env
# Edit .env and add OPENAI_API_KEY and TAVILY_API_KEY
uvicorn app.main:app --reload

API docs: https://researchnet-tsn5.onrender.com/docs

Frontend setup (new terminal)

cd frontend
npm install
Copy-Item .env.example .env
npm run dev

Open the URL Vite prints (normally http://localhost:5173).

How it works

1. Submit a topic using POST /api/research.
2. The backend returns a job ID immediately.
3. The frontend subscribes to GET /api/research/{job_id}/events using Server-Sent Events.
4. Events update each pipeline stage and deliver the completed report and critique.
5. GET /api/research/{job_id} can retrieve the current state.

Notes

• Jobs are held in process memory. Restarting the backend clears them; use Redis/PostgreSQL for durable multi-worker production deployment.
• The current orchestration runs synchronously inside a background task. For production, add a queue (e.g. Celery/Redis or ARQ), rate limits, authentication, and durable job storage.
• URL scraping includes basic protections against non-HTTP URLs and private/local IP targets, but production deployments should add DNS-rebinding-resistant egress controls.
• LLM output and web content are untrusted. Verify citations and claims before relying on reports.