AI product engineering and technical product marketing
Model behavior, product state, APIs, deployment, reliability, and the writing that helps people use what ships.
Products and technical artifacts you can open, inspect, and use.
A self-serve audio-separation product that turns long-running model inference into a flow people can understand, pay for, and recover.
Model orchestration, job state, credits, limits, secure output delivery, deployment, observability, and recovery.
Making asynchronous, variable work legible without exposing the queue or leaving failed jobs as support tickets.
A live model API for coding agents with exact model selection, bounded route attempts, pre-dispatch spend checks, and recorded final charges.
One API across supported models, route health and current pricing data, hard spend controls, and per-request receipts.
Keeping fallback and cost behavior explicit enough that an agent never loses track of model policy, route attempts, or the final charge.
An open local runtime for connected devices, rooms, deterministic scenes, and auditable automation.
I separated the reusable orchestration layer from private Flow HiFi and Flow Music products, then rebuilt the core on a clean TypeScript history with typed capabilities, inspectable plans, scoped execution, idempotency, and exact partial-failure results.
An ocean acoustic intelligence research program that turns long-duration underwater recordings into a reviewable account of what warrants attention and why.
I shaped the review model, public evidence standard, disclosure boundary, and shipped site around authoritative source records, explicit uncertainty, and attributable decisions.
Full-stack AI and machine-learning engineering, from chatbots and agents to predictive systems, computer vision, custom model work, data pipelines, and MLOps.
Product direction, solution architecture, model selection and development, full-stack implementation, integrations, evaluation, deployment, and production operation.
The technical depth has to match the problem: integrate a capable model when sufficient, adapt one for the domain, or train from scratch when the objective requires it.
Infrastructure practice and a complete developer-writing sample.
A self-operated UniFi and Mellanox environment with segmented traffic, Wi-Fi 7, Protect cameras, monitoring, and documented topology.
A TypeScript guide that starts with one request, then adds typed responses, retry boundaries, caching, privacy, and useful errors.
Read the guide