Field notes
Running AI in production, AgentOps, and regulated cloud infrastructure - plus the DevOps foundations underneath.
The hard part is not the model, it is the guardrails around it. Lessons from building Tendwell — deterministic where correctness matters, the propose/approve/execute separation, local-first as a hard default, and audit as a feature.
Field notes on DORA Article 17 in practice — why incident classification is harder than the regulation makes it look, and why the tooling you bring in to help is itself a compliance surface.
The infrastructure and product story behind a fine-grained car classifier — Colab to RunPod, EfficientNet-V2-S at 94.24% on 655 generation-level classes, plate blur, on-device gating, and hierarchical inference.
How I built Terraback — a local-first CLI that turns live AWS, Azure, and GCP infrastructure into clean, modular, importable Terraform that plans zero-diff.
A prototype putting a local LLM agent in front of a Prometheus/Thanos stack. Baseline-aware alerting, a ~4,800-to-2 query batching rewrite, and lessons from a 7B model.
Why DNS-based database failover fails in practice due to connection pools, resolver caching, and TCP behavior, and what solutions actually work.
A practical framework for making infrastructure cost decisions that balance cost, risk, operational overhead, and business constraints.
A framework for evaluating deployment platforms and identifying when complexity has become overengineering. Red flags, questions to ask, and when complexity is justified.
A practical guide to converting your existing cloud resources into maintainable Infrastructure as Code. Covers the challenges, manual approaches, and automated solutions.
A systematic approach to diagnosing and resolving Kong Gateway networking issues in Kubernetes environments. DNS, service discovery, and common pitfalls.
Practical patterns for implementing SSL termination with HAProxy when handling multiple domains. SNI routing, certificate management, and configuration examples.