Together with my co-author Juni (officially my supervisor), I keep this small blog of ideas and discussions. If something here sparks your interest, or you have something to say or add, my DMs are always open. Let's keep it humble and respectful here :)
Deploy your own model with vLLM, connect it from a TypeScript app, and skip the API bills.
The architectural journey from convolutions through ViT to large language models — and what it means for production ML infrastructure.
How we replaced Grafana-Prometheus with OpenTelemetry + ClickHouse for ML workloads.
Lessons learned migrating orchestration at scale.