I build AI-native products that ship, with evals, governance, and auditability
A decade of venture building behind it. Three loops: explore fast, harden what survives, scale what earns.
a validated bet, or a fast kill.
BUILD / DESIGN
PROVE / VALIDATE
EARN / ECONOMICS
I prototype in days, not quarters. Solves the feasibility unknown. Delivers a working artifact showing what the model can and can't do, before anyone writes a spec.
I put rough builds in front of real users early. Solves the assumption risk. Delivers evidence of what people actually want, not what we imagined.
I size demand and willingness to adopt. Solves the nobody-wants-this risk. Delivers a case to fund the bet, or the confidence to kill it cheaply.
Most bets die here. That's the point.
COST AVOIDED
LOOP 02
quality defined, measured, and defensible
BUILD / DESIGN
PROVE / VALIDATE
EARN / ECONOMICS
I turn the prototype into a system. Orchestration, state, checkpointing. Each task routed to the right model on accuracy, latency, and cost. Closes the demo-to-production gap.
I build the eval set from real failures. Names the failure modes we own. Delivers an objective bar and gates that block regressions.
I set adoption and retention targets. Solves the no-definition-of-success risk. Delivers thresholds the product is held to, not hoped for.
Evals come from what broke in Loop 1 not from a whiteboard.
RISK REDUCED, MOAT BUILT
LOOP 03
margin expanded, unit economics proven
BUILD / DESIGN
PROVE / VALIDATE
EARN / ECONOMICS
I optimize architecture and cost structures. Model routing, caching, and fine-tuning. Reduces unit inference cost while maintaining accuracy at scale.
I establish real-time monitoring and feedback loops. Solves latency drift and edge cases. Delivers continuous quality assurance in production.
I align product usage with expansion revenue. Solves margin erosion at volume. Delivers sustainable unit economics and compounding asset value.
Every eval failure at scale is a new Loop 1.