The Intensive
Two focused hours on your operations. You walk away with a written AI roadmap you own. Specific, scoped, opinionated.
Book the Intensive →↳ Rancho Cucamonga · AI Build
~ booked direct, shipped fast, no suits required ~
One problem, one deliverable, fixed price, shipped to a working URL. Working with operators in Rancho Cucamonga and the broader Inland Empire who'd rather hire the engineer than the agency that subcontracts the engineer.
~ three ways to work together ~
Fixed price wherever possible. The deliverable is always something running, not a slide deck about something running.
Two focused hours on your operations. You walk away with a written AI roadmap you own. Specific, scoped, opinionated.
Book the Intensive →One problem, one deliverable. Fixed price, no hourly surprises. Shipped to a working URL with a draft PR you can read end-to-end.
Spec a build →Ongoing ops layer. Cancel anytime. Your AI systems stay running, instrumented, and improving. The kind of arrangement where you sleep better.
Discuss the fit →~ what's on the bench ~
Six artifacts you can read about, fork, run, or talk to right now.
case study · 01
Mood stabilizer for your AI bill.
Local Rust daemon. Polls every provider you pay, normalizes the bill into one SQLite store, answers a single question: what's actually burning this month, fixed and variable, across everything? No SaaS, no telemetry. Just the number.
case study · 02
A reimagined OpenAI Symphony, opinions cranked up.
Fork of OpenAI's Symphony with dual runtimes (Codex + Claude Code), cross-model adversarial pre-PR review, Linear OAuth identity, label-gated autonomous merge, and model-summarized run logs. Same Symphony shape, harness opinions cranked up. v2 in progress.
case study · 03
The tool that builds itself.
Autonomous coding orchestrator. Hand it a markdown spec, planner-generator-evaluator loops until the software exists. Pointed at its own CLI spec: 1,141 tests, four hours, walked away. Pointed at mini-jwt: 38/38 features, 96% coverage, 70 minutes, $9.27.
case study · 04
Tool calls before tool calls had a name.
While working at ACME Autos we shipped one of the first AI-enabled customer support agents, before agents were a thing. Tool calling was hand-coded, pre-harnesses, pre-SDKs, pre-MCP. The case study runs a hardened, period-styled version you can talk to. Try to break it. (You're welcome to. It won't.)
case study · 05
Wind turbine output, mapped over real terrain, in the browser.
Geospatial production-mapping for the National Lab of the Rockies. Full frontend rebuild plus a thin tile-server, deployed at windwatts.nlr.gov. The kind of work where the data has to be right and the UI can't get in the way.
case study · 06
Capture a thought. Ask in natural language. It surfaces what you meant.
Semantic memory layer for AI agents. Postgres + pgvector + an MCP server. Drop a sentence, it gets embedded, classified, and filed. The next agent that needs the context gets it without being told.
~ the engine room ~
What an instrumented operation looks like in practice. Each card answers one question an operator actually has to answer every day. Mocked for this demo, real endpoint shapes underneath.
"what's burning this month?"
month-to-date · projected $632
"what's the harness doing right now?"
running
merged today
"are the workers alive?"
uptime · 4 workers
~ this is what 'instrumented' looks like. yours could too. ~
~ the smith ~
Shawn Petros. Staff engineer, fifteen years deep, operating Petros Industries solo. Shipped for corporate, now shipping for SMB. Building AI tools for smaller teams, on constrained timelines, with no suits between you and the person writing the code.
Based in SoCal. USAF vet. Dad. Stick shift enthusiast. The engineer who never stopped being a creative, and always was a builder.
"Build like a smith.
Ship like a bricklayer."
last 90 days · github
1,549 commits
active
79 / 90
~ proof in the build cadence ~
↳ shipped at
~ pull the cord ~
Tell me what's busted and how soon you need it not-busted. I'll write back with a recommendation, not a slide deck.