shortbulls

What's already built

We run 8 independent AI analyst agents — credit, macro, flow, equity, gold, bond, hedge fund risk, and geopolitical intuition — each scoring the market from a different angle. Their verdicts feed into a weighted consensus engine that produces a composite signal. Credit leads equity by 2-8 weeks, so the credit analyst gets the highest weight (1.4x). When 6 out of 8 agents agree, we run a signal independence check to catch consensus traps — situations where strong agreement actually masks a single shared root cause. The system uses Claude R1 with extended thinking for two modes: live regime classification before market open (06:00 CET), and post-mortem analysis when signals fail — including full agent autopsy to find what went wrong. On the data side, we track COT (Commitments of Traders) for extreme positioning in gold, silver, and crude — when managed money goes >25% net long, that's historically a contrarian sell signal. We monitor BaFin short interest disclosures for the German market and SEC Form 4 insider filings, flagging cluster selling (multiple insiders dumping the same stock) as critical alerts. The finanzmarkt-advisor integration adds chaos theory analysis: Lyapunov exponents measure market sensitivity to initial conditions, Hurst exponents detect trending vs mean-reverting regimes, and permutation entropy filters out noisy signals before they reach the agents. There's also a vision learning agent that extracts chart patterns and price levels from screenshots using Claude Vision.

Who we are looking for

If you trade — even paper trading — and want to build the tools you wish existed, you'll fit right in.

• Developer with Trading Interest — You follow the markets and want to build tools around them. Node.js, Python, or React experience helps, but genuine curiosity matters most.
• Data / Signal Engineer — Extend the real-time pipeline, add new data sources, optimize the agent consensus system.
• Experienced Trader (any level) — Help validate signals, design new alert triggers, and stress-test the system against real market conditions.

Stack

Python, FastAPI, Node.js, React, Supabase, PostgreSQL, Claude AI (R1 + Vision), FRED API, Qdrant, Vercel.