I build AI automations
that survive production.
Most AI automations work in a demo and quietly break under real traffic. I build the ones that don't — durable agent workflows, multi-step automations, and the integrations that connect AI to your real tools.
20+ years building backend systems · Founder of Sire · ex-Zendesk · Go · Python · MCP · n8n / Make
What I can build for you
From a single reliable workflow to a full multi-tenant agent platform — shipped production-ready, not as a fragile demo.
AI workflow automation, production-ready
Multi-step automations and AI agents that connect your tools — APIs, CRMs, n8n / Make — with real error handling and retries. A working, documented system you can rely on.
Turn an AI prototype into a real product
Have a ChatGPT or LangChain demo that isn't production-ready? I re-architect it for reliability: state, retries, logging, and deployment — so it survives real users.
Agent orchestration & platform builds
Need durable, multi-tenant agent orchestration built from scratch? I've built exactly that — and can architect and build yours, avoiding the demo-to-production traps.
Integrations & MCP
Connect AI agents to your real systems via the Model Context Protocol and custom API integrations — so agents act on the world, not just chat about it.
Selected work
Things I've built and shipped. Most are public — click through and see the code.
Sire ↗
A platform for building and running autonomous AI agents as durable workflows — surviving crashes, enforcing human-in-the-loop safety gates, and connecting to your tools via MCP. Multi-tenant, Go + Next.js.
Zerfoo ↗
A from-scratch ML inference framework in pure Go (zero CGo) that runs LLMs faster than Ollama, supports 41 model architectures, and serves models larger than RAM via memory-mapped I/O. Proof I understand AI at the metal.
Mint ↗
An open-source CLI that converts any OpenAPI 3.x spec into a working MCP server — discover, generate, publish, and deploy with one command. The bridge between AI agents and real APIs.
Gist ↗
A Go library that indexes content and retrieves the most relevant snippets within a token budget (stemming → trigram → fuzzy). The context/RAG plumbing that makes AI apps accurate and cheap to run.
Wolf
An end-to-end autonomous decision system: Go microservices over NATS, a JWT-secured gateway, a real-time ingest → strategy → risk → execution pipeline, and on-device LLM inference — plus a SwiftUI companion app.
Spark ↗
A Podman-backed, Kubernetes-style orchestrator for running GPU/ML workloads on a single machine — submit standard manifests and get tracked, logged, reproducible runs.
Experience
Two decades shipping backend systems — the last year building AI agent infrastructure full-time.