Rust · Decision Systems · Causal ML
I'm building Calybris, a decision engine in Rust that logs why it chose what it chose. The open-source core is on GitHub.
My focus is on systems where decisions need to be traceable and reproducible — AI model routing, cost governance, analytical pipelines. Still early stage, learning in public.
$ _
Proof-carrying, domain-neutral prescriptive decision engine. Rust. Every decision gets a fingerprint, cost estimate, risk penalty, quality floor, and replayable audit trail.
AI cost governance gateway. Every LLM call must justify its cost. Routes, downgrades, blocks, caches across OpenAI · Anthropic · Google based on expected value, risk, confidence, and tenant budget.
Quant signal evidence audit layer. Not a trading bot — a falsification engine. Tests whether signals survive cost, leakage, overfit, walk-forward, and complexity pressure before capital goes in. MDL-inspired proxy: more complex signal requires stronger evidence.
AI finance reporting for SMEs. 3-agent pipeline (Planner → Executor → Critic), Holt forecasting, anomaly detection, incentive matching, PDF generation. Google Gemini Flash.
Prescriptive churn engine. Prediction → SHAP explanation → DiCE counterfactual prescription → DoWhy causal analysis → NPV decisioning → T-Learner CATE → Backtesting.
7M-row benchmark: S-Learner, T-Learner, X-Learner, DR-Learner, Causal Forest. S-Learner won.
SHAZAO1AST7J35E81C28E1A8YJ7DMHD90WRSS4QYUCTATVLYcaXP3ioKZp2zsmdCYOpen to feedback, collaborations, and conversations about decision infrastructure.