I started by proving theorems,Now I build systems that learn.

Showing featured projects. More work lives below the fold if you have brought coffee.

1. 01Does momentum help or hurt when the objective keeps moving?Accelerated PPO Variants · 2026Momentum helps PPO only when it acts on the policy that generates the data, and only where that policy sequence has a consistent direction.Read
2. 02Can you build an agent where a made-up number fails to typecheck?Numbers with Receipts · 2026The guarantee is not statistical: an answered result that isn't grounded cannot be constructed.Read
3. 03Can a policy learn from logged data without ever exploring?Offline RL for Autonomous Driving · 2025Stable driving can be recovered from logged trajectories alone, in minutes instead of hours of interaction, but only as far as the inherited dataset allows.Read
4. 04What does it take to make federated learning trustworthy end to end?Federated Learning as a Lifecycle · 2025Trustworthiness is a lifecycle property, not a training-loop one: once every phase transition has to pass an executable policy check, auditability stops resting on informal operational assumptions.Read
5. 05How do you catch and explain pension anomalies without touching the payroll engine?Retirement Pension Anomaly Detection and Causal Diagnosis · 2025The payroll engine's outputs alone were enough to flag anomalies at 2.5% reconstruction error, but it was the causal attribution layer that turned those alerts into something an auditor could act on.Read
6. 06Can countries build shared flood intelligence without centralizing their data?Federated Flood Prediction for Climate Adaptation · 2025Countries can contribute to a shared continental flood model without surrendering their own hydrological data, which makes the obstacle to continental forecasting institutional rather than technical.Read

Recognition