WARNING: NOT AN INVESTMENT TOOL. This is a technical demonstration of multi-agent AI systems. All output is AI-generated and must not be used as financial advice.
System Architecture

ProspectAI Pipeline

A 6-agent CrewAI flow that runs market analysis, technical and fundamental research in parallel, drafts a portfolio strategy, stress-tests it with an adversarial critic, then produces a revised final report. Select a sector and run it live to watch the pipeline animate.

The diagram below animates automatically to show the execution flow. Run a real analysis from the Home page.

Phase 1
Market Analyst
RedditSentimentTool SerperDevTool Haiku
Phase 2 — Parallel
Technical Analyst
TechnicalAnalysisTool yfinance Haiku
Phase 3 — Parallel
Fundamental Analyst
FundamentalDataTool FundamentalGraderTool Haiku
Phase 4
Draft Strategist
CompositeScoreTool PortfolioAllocatorTool Sonnet
Phase 5 — Adversarial Quality Gate
Critic
Reasoning only Sonnet
Phase 6
Final Strategist
PortfolioAllocatorTool Sonnet

Tech Stack

Orchestration
CrewAI Flow
Multi-agent pipeline with parallel phase execution and structured Pydantic output validation at each step.
LLM
Anthropic Claude
Haiku for data-gathering phases (1–3), Sonnet for reasoning phases (4–6). Per-agent model override via env vars.
Market Data
yfinance + Reddit
OHLCV price history, fundamentals, and sector sentiment from Reddit (PRAW) with SerperDev fallback.
Backend
Modal + FastAPI
Serverless GPU-optional container. Streams agent progress via SSE. Persists analytics and LONG-BUY signals to Modal Dicts.
Report Storage
Cloudflare R2
Pipeline results stored as JSON (YYYY/MM/DD/run_id.json) in a private R2 bucket. Served via backend proxy.
Frontend
Cloudflare Pages
Static SPA built with Vite. No framework — vanilla ES modules. Real-time updates via EventSource.