I work on the parts of software that get interesting when they fail: resumable agent workflows, observable services, and protocol-heavy systems.
Why these problems?
Most software looks reliable on the happy path. I care about what happens after interruption: whether state can resume, whether a failure leaves useful evidence, and whether protocol code makes every byte explainable.
- SignalOps — a self-hostable Kotlin backend that turns telemetry into inspectable incident evidence, timelines, confidence, and limitations.
- open_loop_service_handler and DesfireCrypto — C++ tooling for typed transit-card data and DESFire cryptography.
Merged: Kotlin #6982 — clearer browser-runner failures · Google ADK for Go #1247 — streaming confirmation coverage
In review: Genkit proposals · Koog #2184 · vinext #2829