I build and ship production AI systems, end to end, on a team or on my own.

I started building Goalvana the last week of June 2025. The first production version was live about six weeks later. I had beta testers on it inside four months. In under a year, alone, I have taken it to eight deployable services, 336 API endpoints, 168 data models, 274 migrations, and over a thousand commits.

It is a coaching operating system: one platform to build, deliver, and grow a coaching practice, replacing the five-plus disconnected tools coaches stitch together for calendars, payments, programs, and messaging.

The engineering I am proudest of:

• A program builder that turns a plain-language brief into a full structured curriculum, run by an orchestrator and parallel workers with a deterministic executor.
• A support agent built on a real tool-use loop, no framework, with 16 tools that read and write live data, capped and guarded so it cannot run away.
• A multi-provider LLM stack (Anthropic, OpenAI, ElevenLabs, Deepgram) with prompt caching, response caching, and per-user token and quota tracking.
• A database-driven prompt system with four-level override, so non-engineers can tune AI behavior without touching code.
• The full product around all of it: React and Next.js front ends, Fastify and TypeScript services, Prisma and PostgreSQL, Stripe Connect for payments, Google Calendar OAuth, RRULE scheduling, Daily.co video, and Sentry, on Railway and Vercel. I held the quality line solo with 711 backend tests and 763 end-to-end tests across 89 spec files.

Real-time voice interview practice. You talk out loud with an AI interviewer that talks back, holds a real conversation, and scores every answer on structure, specificity, and impact against a 0 to 100 readiness model. It mines your best STAR and CARL stories through guided conversation, then feeds you weakness-weighted drills across multiple professional tracks.

The hard part is the voice loop: a real-time, full-duplex pipeline on Pipecat and Daily.co WebRTC, with Deepgram transcription, Silero voice activity detection, GPT-4.1 driving the conversation, ElevenLabs on speech, and GPT-4o handling the scoring. It runs on the same program engine I built for Goalvana.

Real-time scoring, ticketing, and fundraising for judged competitions, built solo in about four months across 200-plus API routes. I scoped one customer domain's exact workflow and shipped a configurable platform around it, which widened the market well past the first use case.

• Multiple judges score live over Server-Sent Events, synced across audio, video, and scores on a single clock, with scoring scales that adapt to any judged event.
• An AI pipeline transcribes judge audio with Deepgram and writes performance and event summaries with Claude, exported to PDF.
• Stripe Connect handles payments end to end: Express accounts, a 3% application fee, QR ticketing with tiered pricing, and goal-tracked fundraising.
• Cross-product single sign-on verifies Goalvana-signed tokens, so users move between both platforms on one login.

I take on a small number of advisory and due-diligence engagements. I embed with companies to scope technical strategy, architecture, and delivery against real constraints, then own the result. I also run technical due diligence on both the buy side and the sell side, assessing architecture, risk, and team for investors and acquirers. I have been on both sides of an acquisition, so I know what the other side is looking for.