./builds

What I've built.

Selected work — own products and client projects.

NDA Project

Live Financial Literacy Platform

NDA
  • Next.js
  • TypeScript
  • Supabase
  • Realtime
  • Row Level Security
  • AI Avatar
  • OpenAI
  • Vercel
  • Sentry

Role

Senior Architect & Solo Developer

The Challenge

A financial education program needed a live, group-based simulation for workshops with 30–50 simultaneous participants. The room votes collectively on money decisions, but the financial consequence has to be calculated per person from their own balance — so the same vote lands differently depending on where each participant started. Anything that breaks does so in front of a live audience, on a shared screen, with a non-technical facilitator running the room under time pressure.

The Approach

Chose Next.js and Supabase over no-code tooling for precise latency control and pixel-accurate large-screen layout. Built a concurrency-safe state engine using atomic guarded updates, closing a time-of-check-to-time-of-use race condition and adding idempotency protection so a re-evaluated round can never double-apply financial effects. The AI avatar layer reconnects per reaction rather than holding a persistent session, avoiding idle-time cost and connection races. Content and simulation logic are fully separated, so the entire narrative layer swaps via a single toggle without touching the engine underneath.

Timeline

Under 2 months

Outcome

In production since July 2026 with a real cohort, running weekly sessions across a six-to-eight week curriculum. A mass-lock incident in the first live session was root-caused from production data within minutes and resolved.

Full-cycle build — architecture, planning & complete implementation.

NDA Project

Live Financial Literacy Platform

NDA
  • Next.js
  • TypeScript
  • Supabase
  • Realtime
  • Row Level Security
  • AI Avatar
  • OpenAI
  • Vercel
  • Sentry

Role

Senior Architect & Solo Developer

The Challenge

A financial education program needed a live, group-based simulation for workshops with 30–50 simultaneous participants. The room votes collectively on money decisions, but the financial consequence has to be calculated per person from their own balance — so the same vote lands differently depending on where each participant started. Anything that breaks does so in front of a live audience, on a shared screen, with a non-technical facilitator running the room under time pressure.

The Approach

Chose Next.js and Supabase over no-code tooling for precise latency control and pixel-accurate large-screen layout. Built a concurrency-safe state engine using atomic guarded updates, closing a time-of-check-to-time-of-use race condition and adding idempotency protection so a re-evaluated round can never double-apply financial effects. The AI avatar layer reconnects per reaction rather than holding a persistent session, avoiding idle-time cost and connection races. Content and simulation logic are fully separated, so the entire narrative layer swaps via a single toggle without touching the engine underneath.

Timeline

Under 2 months

Outcome

In production since July 2026 with a real cohort, running weekly sessions across a six-to-eight week curriculum. A mass-lock incident in the first live session was root-caused from production data within minutes and resolved.

Full-cycle build — architecture, planning & complete implementation.