AI that gets used, not just demoed.
I ship LLM systems into real business workflows — from messy Japanese documents to evaluated, production-ready pipelines. A decade of product design and engineering is why people actually adopt them.
LLM pipelines shipped for 2 client businesses
Eval harness · 146 labelled attacks scored
10 yrs product · 3+ yrs frontend leadership
EN · 日本語 · 中文
Selected projects
Strobe Assistant—Extending Strobe with an AI Agent
How I extended Strobe, a photo-sharing app on AWS, with a conversational AI agent, an asynchronous pipeline that makes every photo searchable, and an agent that runs on its own every morning, plus what I'd do differently.
Understanding the Log Types in a Wazuh + Claude Security Pipeline
A field guide to the five log types flowing through my IFN737 research pipeline — from raw attack execution logs to Wazuh alerts to Claude's structured classifications — and how they connect.
LingoBun: Hardening an AI Language-Learning Product for Production
An LLM-powered vocabulary app taken past the demo stage: Redis-backed rate limiting and cost caps on every AI call, Zod-validated environment config, and a 180-test Vitest suite guarding the pipeline.
AuStride: Striding Towards an Active Future
Walking or riding rather than driving offers the two-fold benefit of increasing mental and physical health while reducing motorised transport’s impact on the environment. However, there are several challenges to consider: availability of safe and accessible walking and cycling options, lack of time and motivation, and gaps in transport infrastructure that make it impractical to walk, ride, or use public transport.
How I work
Scope it with the customer
Start from the workflow, not the model — where AI can be trusted, where it can't, and what a human has to sign off on.
Build the pipeline
Retrieval, orchestration, guardrails, cost and latency budgets — production code, not a notebook demo.
Prove it with evals
Ground truth, a scoring rubric, and a re-run before every change — so "it works" is a number, not a feeling.
Posts
Strobe Assistant Part 1—AWS Target Groups and Application Load Balancer (ALB)
What an Application Load Balancer's listeners, rules and target groups each do, how health checks de...