Work
Things I built because the tool did not exist. Each page states the problem, the design decisions and why, and what I would do differently — including the tradeoffs I am not happy about.
On what is not here. Some of my work is done inside AWS on customer cases, and none of that can appear on a public site — no code, no customer names, no case detail. Where a project came out of that context I describe the architecture and the reasoning only. If you want the depth behind it, ask me in a conversation where I can answer properly.
-
A multi-agent pipeline for technical diagnosis
A script holds the control flow; the agents do the thinking. Redundant parallel investigation, then reconciliation, then a build-then-independently-verify gate before a word gets written.
-
Offline hybrid search over a personal case archive
Semantic similarity and SQLite full-text search, fused by reciprocal rank. Local ONNX embeddings, two dependencies, no network calls at all. 554 lines.
-
A presentation engine that checks its own layout
Slides as a plain data array, no framework, and an in-page probe that measures every slide against the real obstacle and reports overflow to a headless browser.