Architecting decision intelligence
for AI-native enterprises.
Fourteen years shipping enterprise products. Two consecutive 0-to-1 enterprise AI platforms. Now operating the decision layer between LLMs and the enterprises that depend on them.
01Selected work
Live LLM products I ship and operate. The patterns I learn here show up in enterprise work weeks later.
Operator-Builder
Facet (usefacet.app)
LLM-powered web app that rewrites LinkedIn profile content for tech workers between jobs. Free, no signup. 100+ sessions in the first 48 hours from organic distribution.
Operator-Builder
AI Crew Chief
Compound AI system that decides go/no-go between motorcycle racetrack sessions. Four-layer architecture with the LLM as orchestrator, not decider. Built as a personal project; the patterns now inform AI-native decision intelligence work at SAS.
02About
Built like a product, run like an operating system.
I treat AI products as probabilistic systems, not feature trees.
Eval-driven roadmaps replace opinion-driven ones. Acceptance criteria become measurable; the “is this good enough yet?” question becomes the PM’s job.
- AI-native SaaS, decision intelligence, AI/ML asset governance, LLM in production.
- Two consecutive 0-to-1 platforms shipped. Concept to GA with end-to-end ownership.
- Compound AI over single-model. The LLM is rarely the decider.
End-to-end ownership is a strength, not a constraint.
I ship live LLM products on the side. The patterns I learn there show up in enterprise work weeks later. Concept to GA across product, GTM, partner enablement, and analyst relations without dedicated PMM.
- Anthropic, Gemini, Llama via Groq. ChromaDB, MLflow, Microsoft Fabric.
- Write it down. If a decision isn’t in a doc, it didn’t happen.
- Real authority on what the product becomes. Not just what ships.
03Writing
Notes on AI strategy, decision intelligence, and the market underneath them.
- May 15, 2026
We're Building Agents Wrong
Devin shipped Interactive Planning a year ago. The most-hyped autonomous AI agent rebuilt itself around a human checkpoint. The industry still designs new agents as if it didn't happen. Here's where AI PMs should be designing instead.
- May 09, 2026
Beyond Chatbots: Why I Built a Compound AI System to Trust My Life at 160mph
The first AI bot I built told me my brakes were safe. They weren't. Here's why I rebuilt it as a compound AI system, and what that means for any team building production AI.
04Recruiter Copilot
Paste a JD. Get a grounded fit assessment in seconds.
Job description analyzer
Streams a Claude-generated fit analysis grounded in my résumé, philosophy, and project highlights. Four sections: fit score, top strengths, gaps, talking points.
OutputFit Score · Top Strengths · Gaps · Ask About. Streamed and grounded in my public material only.