About Me



I'm Guillermo. I co-founded Levee as CPO and built its tech stack, and I now act as CTO and sole engineer across six more company platforms. The unusual part isn't the count—it's that on every one of them, the person making the product call, writing the code, and talking to the customer is the same person.
How I actually work
Nothing gets handed off. Most teams lose weeks in translation: a stakeholder describes a problem, a PM writes it down wrong, an engineer builds what was written. I collapse that into one loop. I sit in the room, hear the real constraint, and it's in production that week—because I never had to explain it to anyone.
Agentic coding is the multiplier, not the author. I run a fleet of coding agents in parallel worktrees, but the leverage comes from the discipline around them: a written spec before any code, TDD lanes, and adversarial review gates where a second agent's job is to disprove the first. Unreviewed agent output is a liability, so I treat it like a junior engineer's PR—every diff gets read. That's how I landed 10.5K commits in six months without shipping slop.
I migrate things while they're running. Two live platforms moved onto new infrastructure last year without pausing the roadmap: Levee off AWS and Terraform onto GCP with Kubernetes and Flux GitOps, and SunrAI off Supabase onto AlloyDB behind self-hosted PostgREST with Stack Auth. Both are boring now, which was the point.
Technical depth
Scale, as the only engineer. SunrAI is 637 API routes, 602 migrations, 490 components, and 1,606 test files, with a migration ledger that enforces exact staging checksums before anything touches production. Levee spans 15 repositories—Django domain API with an embedded Rust service for latency-critical paths, FastAPI agent orchestration, a native Swift/SwiftUI iOS client with ARKit, and an MCP server that exposes the platform to AI agents.
AI in production, not in a demo. Computer vision (YOLO and RT-DETR) inspecting real hotel rooms at a 605-room Marriott pilot, where it cut inspection time 60%. Multi-model LLM orchestration with a cost-aware model registry. Agents that run on a schedule with nobody watching—and the idempotency, cursor discipline, and abort-don't-skip semantics that make that safe.
Money-moving code. Stripe Connect onboarding, prize pools with charity splits, platform fees, sales tax, payouts, and refund and dispute state machines. Plus 18 Solidity contracts on Base L2 with 774 passing Foundry tests. When code touches money, the tests stop being optional.
Before all this
I came up through product, not a CS program, which is why I'm comfortable in a stakeholder meeting and a Kubernetes manifest on the same afternoon. I managed a $250M+ IoT portfolio at Chamberlain Group and turned the Ring partnership from −11% to +68% IRR. I hold a B.S. in Neurobiology from the University of Iowa and PSPO credentials, and I teach product management and SQL at UCLA—which is excellent practice at explaining hard things to smart people.