Where this stands
Case study in progress. Target completion: 2026-08-15. The full writeup lands once the research, build, and eval phases are done.
The customer scenario
Hedge fund deploying LLM-driven trading agents needs production-grade guardrails, kill-switch monitoring, and regime-aware risk management.
My product canvas
This case study is grounded in WealthPilot — an autonomous self-improving trading system built on Python + Alpaca, running 10 edge tools across signals (congressional trades, options flow, dark pool, Reddit sentiment) into a composite conviction score. The hardest engineering problem isn't generating signals — it's building a system honest enough to recognize when its signals have stopped working. Canonical writeup: kaydenlabs.com/work/wealthpilot.
Architectural patterns to be demonstrated
- 10-layer risk control architecture — Kelly position sizing, ATR stops, drawdown circuit breaker, regime-adjusted sizing, correlation penalties, earnings blackout windows
- Regime-aware autonomous kill-switch — the Plutus agent has halt authority on detected regime shifts, not just on drawdown. Halted live in March 2026 on a 0% win rate in a choppy regime.
- Weekly walk-forward retrain with Monte Carlo stress validation — logistic regression on 436 samples with time-decay weighting, validated against COVID, 2022 rate cycle, and SVB scenarios
- Trade-log signal-weight adaptation as a self-correcting eval loop — 497 automated tests across 32 files
What you'll find here when it ships
- Architecture diagram
- Eval set with results
- Cost and latency analysis
- Failure modes documented
- "What I'd do differently" retro