I'm a full-stack AI engineer: I build entire platforms end to end, with AI as the load-bearing part — turning models into shipped products rather than demos. Lately my attention is on small language models and agents, and the question underneath them: can you actually trust what they produce?
That question pulls at me more than any benchmark. I build with small language models, wire up agents with guardrails I'd stake something on, and keep circling back to verification — proving an answer is faithful to its source before anything depends on it. I care less about a model that demos well than about output you could hand to someone without flinching.
There's a quieter thread, too: security and reliability. I like understanding how things break, and building the unglamorous layer that keeps them from breaking silently. I don't write much about that work — but the habit of mind shows up in everything else I make.
What's actually public.
SpotifyScraper is the honest headline — a Python library that pulls clean data out of Spotify's web player without official API keys, and the one piece of my work real people actually use. Most of the rest lives in private repos, client projects, or the experiments pile, and I'd rather under-claim than dress it up. The numbers below are real and fetched live. That's the whole proof.
A short arc.
The one that shipped
SpotifyScraper
A Python library that pulls clean data out of Spotify's web player without official API keys. It found real users and a small community — still the piece of my work the most people have actually touched.
The wider range
Platforms, tools, and experiments
Full-stack web platforms, AI agents and tooling, automation, the occasional native Apple app. Much of it is private or unfinished — and the unfinished work is where I learn the most.
Now
AI, mostly — and trusting it
Small language models, agents with real guardrails, and a quieter obsession: proving model output is faithful to its source before anything is allowed to depend on it.
