Engineer. Pilot.
Builder.
I build production software and the AI systems around it, end to end: full-stack applications, data pipelines, and multi-model LLM workflows. Founder of XPRT Solutions LLC. Open to senior and founding engineer roles.
What I build
Full-stack applications
End-to-end products: data model, backend, API, and UI. Python and TypeScript, FastAPI and Next.js, PostgreSQL and pgvector. Owned solo from schema to ship.
AI systems
Multi-model LLM orchestration, retrieval-augmented generation, structured output, and evaluation. AI as a co-developer, used with intent, not as a gimmick.
Data pipelines and infra
Event-driven integrations, webhooks, idempotency and dead-letter handling, GPU processing, and self-hosted infrastructure on Docker and Cloudflare.
Selected work.
Shipped systems, honest writeups.
XPRT Cloud
A drone-mapping SaaS built solo end to end: flight planning, AI photo analysis, GPU photogrammetry, 3D viewer, Stripe billing.
Conversation Intelligence
An AI system I designed: transcript ingestion to structured LLM analysis to retrieval-grounded coaching chat with citations.
Quantum QAOA on IBM Hardware
Ran a selection problem as a QAOA circuit on a real 156-qubit IBM processor. Honest result, real finding about constraint encoding.
TradingView to FreqTrade Bridge
A production webhook bridge translating strategy signals into exchange orders sub-second, with auth, allow-listing, and clean failure modes.
How I build with AI
AI as a co-developer, used with intent.
Evaluating and directing AI output is a core, daily discipline for me, not an afterthought. This is how I ship faster without shipping slop.
Model selection by task profile
I route work across GPT-5, Claude, Gemini, Grok, and local Ollama by what each does best: vision, structured output, code review, multi-agent reasoning. Constant comparative evaluation of where each model is right and where it fails.
AI-gated code review
I ship production code through an AI review gate, reading model output critically and correcting it before merge. On a strong day that is 11+ pull requests, all reviewed, all green.
Structured output, validated
Strict JSON-schema output validated with Pydantic and Zod, so I catch and characterize exactly where a model violates a schema, omits a field, or hallucinates.
Prompts plus iteration loops
Domain-specific prompt design with deliberate iteration is my normal workflow, codified into a persistent skills-and-memory system that carries context across sessions.
Open to senior and founding engineer roles.
Remote-first, US-based, US citizen. I respond within 24 hours.