I build AI products end to end. Focused on LLM-powered web apps, and production ready websites for real clients.
A browser-based battery test data analyzer — built in a weekend, zero dependencies. Keep scrolling to walk through how it works.
A synthetic Li-ion generator (tanh OCV curves, IR drop, per-cycle fade) or CSV upload — capacity computed by trapezoidal current integration, entirely client-side.
Robust z-scores (median/MAD) on fade slopes, sudden capacity steps, thermal outliers, and voltage sag vs fleet. Healthy cells trace slate — failing cells trace red.
Run stats compress to JSON and one BYO-key request returns a test-engineer summary — fleet health, outlier cells, recommended next tests.
The battery test analyzer from the walkthrough above — synthetic Li-ion cycling data or CSV ingest, four statistical anomaly detectors, and an optional Claude-drafted engineering summary. Zero dependencies, hand-rolled canvas charts, fully client-side.
AI chart-analysis tool: a raw chart screenshot becomes a structured technical read — trend, key levels, and two probability-weighted scenarios — in ~12s. Vision prompts with strict JSON schemas render straight into UI components, plus a retrieval-grounded coach that reasons only over the user's logged trades.
The chat bubble in the corner is a real product: a hosted API that answers recruiter questions about my background using an LLM grounded in my resume and project docs. Try it — it's the fastest way to screen me.
A Manifest V3 extension that explains any highlighted text in place — built, debugged, and published to the Chrome Web Store end to end. Streaming SSE architecture, Shadow DOM isolation, and BYOK auth keeping it fully client-side.
Production sites for small-business clients in the automotive and apparel sectors — 38Carmodz and BM Wrapz — hand-rolled HTML/CSS/JS with no framework overhead, owning scoping, build, and delivery end to end.
The layer I build products around — every project on this page ships an LLM in production, not a demo.
From product scripting down to assembly — comfortable at every level of abstraction.
Full-stack delivery: React UIs when the product calls for it, hand-rolled vanilla when it doesn't.
The unusual part of my stack for a web person: real systems coursework — device drivers, file systems, OS internals.
Built and shipped production sites for small-business clients in the automotive and apparel sectors, owning scoping, build, and delivery end to end. Developed an AI-assisted workflow using Cursor and Claude to cut build time, and managed client relationships directly — translating non-technical requirements into scoped deliverables.
