Full-Stack Engineer · AI & Automation
I work across React/Next.js, Python, and .NET — and ship the hard parts: data pipelines, cloud, and real UX.
Flagship · production ecosystem
I grew a fragile internal toolset into 10+ production apps used daily by 30+ people, reclaiming ~1,000+ engineering hours a year.
Migrated a legacy Access database (12 tables, no referential integrity) to a normalized SQL Server schema and rebuilt it as a WPF / .NET 9 app on Clean Architecture + EF Core — reclaiming ~733 hrs/yr (~$41K) for 30+ daily users, synced to the Ajera ERP via ODBC.
A Python / FastAPI + Next.js platform fusing the Ajera ERP (ODBC) and Vantagepoint CRM (REST / OAuth2) into 3 forecast models and a 12-month KPI dashboard — ~90% effort reduction versus the manual Excel process it replaced.
A 4-server MCP stack for AI-assisted development (Claude Code + CLAUDE.md enforcement), the firm’s Revit add-in suite architected across 4 versions, and 7 authored SOP/standards docs.
Signature project
Auto-playing the real pipeline — the orb’s state lights the matching stage. Tap “15 skills” to expand.
Say a wake word and talk to it: it transcribes your speech with Whisper, thinks with a local Ollama model (or a cloud LLM), streams a spoken reply back in under a second, and calls tools to actually get things done on the computer.
Runs locally by design — the repo is private; the architecture write-up below stands in for the code.
Portfolio
A through-line of real products — most of them thread AI into something people actually use.
Live demo AI-integratedA Stack-Overflow-style Q&A platform: post a question and an LLM auto-drafts an answer alongside community responses.
What's hard: Genuinely deployed and clickable — real auth, Next.js, live on Vercel.
Upload a song or video and it renders a 3D brain that lights up and moves to the music in real time — predicted by Meta’s TRIBEv2 brain-encoding model on an on-demand cloud GPU.
What's hard: Production AI-infra: GPU inference at ~$0.02/clip that scales to zero, 3 swappable model backends behind one contract, and a Postgres + Redis queue + worker for load.
Live demoA focused, responsive marketing page built as a front-end craft sample.
What's hard: Pure responsive-UI craft — framed as exactly that, nothing more.
Team project · my piece: Luna AI-integratedCross-platform iOS + Android wellness app with peer-to-peer messaging and activity tracking. I built Luna — a ChatGPT-powered companion — solo.
What's hard: Luna threads OpenAI through a C#/Xamarin MVVM app; the Google Places finder, P2P messaging, and cross-platform build are the parts that genuinely shipped.
Practice projectEvent-driven forex bot trading a practice OANDA account with Bollinger-band mean-reversion, risk-based position sizing, and a React dashboard.
What's hard: Risk-sized positions across an ~80-pair engine; the real stack is Python / pandas / OANDA, not the old tag-soup.
