The case studies below show how I approach distinct kinds of problems in AI engineering. They are not redacted client engagements, they are demonstration work that exists publicly so the methodology can be evaluated before a conversation starts. Each card links to the full case study with the problem, approach, and current status. Source repositories are linked from there.

AIOrchBuilder

Open source

Multi-agent orchestration framework that builds applications by routing specialist tasks to the language model best suited for each kind of work.

Seven specialist agents with defined contracts coordinated by an orchestrator that dynamically routes each task to the right model. Includes Ollama support for self-hosted local models, RBAC wired in from Day 1, and an auditable retrospective path for every agent action.

  • Next.js
  • TypeScript
  • React
  • Python
  • Supabase
  • PostgreSQL + RLS
  • +4 more
Read case study

ClaudeMCP

Open source

Local multi-backend LLM gateway. One process speaks four backends through three API protocols, so existing client code works unchanged.

Single local gateway process serving Claude, Gemini, LM Studio, and Ollama through Anthropic, Gemini, and OpenAI API protocols. CLI auth reuse eliminates double billing for Claude Max and Gemini subscribers. Multi-instance support for local backends.

  • Node.js 20+
  • TypeScript
  • Express
  • Zod
  • Vitest
  • SQLite
  • +4 more
Read case study

FieldForce Tool Tracker

Commercial

Real-time asset intelligence platform for utility and construction operations.

Unified asset management platform combining sub-second GPS tracking, PostGIS geofencing, QR/barcode custody management, fleet health, and bi-directional ERP integration. Three apps in a pnpm + turborepo monorepo: NestJS API, React 18 web dashboard, Expo mobile app with SQLite offline. Code preserved for commercial deployment.

  • Node.js 20
  • NestJS
  • TypeORM
  • React 18
  • Vite
  • MUI v6
  • +12 more
Read case study

Also building

Three additional public repositories that round out the work. Research papers and archives rather than full case studies. Each links to its own README.