Build stories from the trenches
How we actually build and run production systems — AI platforms, Kubernetes, autonomous agents. Real architecture, real trade-offs, no fluff.
Building a BlackRock-Aladdin-style intelligence platform with a free-LLM cascade
A medallion data architecture, a tiered-LLM cascade that routes grunt work to free models and reserves the frontier model for decisions, and credibility scoring grounded in academic truth-discovery research. 28,000+ signals processed — here's how it's wired.
I pointed autonomous AI agents at my own repos — they shipped verified PRs
An audit → fix → verify-gate → PR workflow topology, with honesty guardrails that caught fabricated review markup on a live site. What worked, what needed a human, and why the verify gate is the whole game.
Running a multi-environment client platform on a single k3s box
Dev, staging, and production isolation for a real booking platform — sealed secrets, automatic TLS, private CI, and a shared-services layer — all on lightweight Kubernetes. The setup and the trade-offs of going lean.
Building something that needs to actually work in production?
Let's talk →