I like infrastructure that stays boring under pressure: systems that explain their own failures, recover without a human, and don't surprise the people on call. I mostly write Go, and most of my work goes upstream, where one fix reaches everyone who runs the tool.
40 pull requests merged across 5 projects: Podman Desktop (25) · Grafana (9) · OpenTelemetry Collector (3) · AWS CDK (2) · Kubernetes (1)
See all merged · 15 more in review, including 5 in Kubernetes
Highlights
- Podman Desktop — telemetry now turns itself off when it detects a CI environment, so pipelines no longer have to disable it by hand (#19266).
- Grafana — fixed panels rendering blank when the legend display mode is hidden (#134866).
- OpenTelemetry Collector — component authors can declare feature gates in
metadata.yamlandmdatagengenerates their documentation (#14130).
Recently merged
- Kubernetes — test/e2e/node: use CUDA base image for GPU sanity test on amd64 · Oct 2026
- Grafana — TimeSeries: Fix stacked series not restacking when the data changes · Oct 2026
- Grafana — PanelEditNext: Fix transformation editors ignoring the transformation's own filter · Oct 2026
- Raft KV store with deterministic simulation testing (Go) — a linearizable key/value store on Raft I wrote from the paper. A deterministic simulator injects partitions and crashes and can replay any failure from a single seed. Across 10,000 seeds it found 4 real consensus bugs.
- eBPF observability agent (Go) — HTTP/HTTPS request tracing with no changes to the traced app. It hooks the kernel and TLS libraries (OpenSSL, Go
crypto/tls) and exports OpenTelemetry traces and metrics. Adds +0.65 ms p50 latency at 1,000 req/s, and one binary runs on kernels 5.15 and 6.8.
| Languages | Go, Python, TypeScript |
| Containers & orchestration | Kubernetes, Docker, Helm |
| Cloud, IaC & CI/CD | AWS, Terraform, GitHub Actions, ArgoCD |
| Observability | OpenTelemetry, Prometheus, Grafana, Jaeger |
| Systems | Linux, eBPF, gRPC |
| Data | PostgreSQL, Redis, Kafka |



