QuantDesk

Multi-agent financial research assistant. Ingests SEC filings, earnings transcripts, and market data; dispatches ratio analysis, sentiment scoring, and peer comparison to specialized LangGraph agents; grounds every response via hybrid RAG (BM25 + Qdrant). Cuts a documented analyst workflow from ~4 hours to ~25 minutes.

RAGAS eval on 200 questions: faithfulness 0.91, answer relevancy 0.88

---

Stack

• Agents: LangGraph
• LLM: Claude (Anthropic API)
• Vector store: Qdrant
• Observability: LangSmith
• API: FastAPI
• Infra: GCP Cloud Run + GCS
• Frontend: React + TypeScript + Vite + Tailwind

---

Project structure

quantdesk/
├── backend/
│   ├── agents/          # LangGraph agent definitions
│   ├── api/             # FastAPI routes and middleware
│   ├── core/            # State schema, config, types
│   ├── data_ingestion/  # SEC EDGAR + transcript ingestion
│   ├── rag/             # Hybrid retrieval (BM25 + Qdrant)
│   ├── evaluation/      # RAGAS evaluation harness
│   └── utils/           # Shared utilities
├── frontend/
│   └── src/
│       ├── components/  # UI components
│       ├── hooks/       # Data fetching hooks
│       ├── pages/       # Route-level pages
│       └── store/       # Zustand state
├── infrastructure/
│   ├── gcp/             # Cloud Run + GCS Terraform
│   └── docker/          # Dockerfiles
├── PRD.md
└── docker-compose.yml