Owner of an end-to-end consumer-scale probabilistic forecasting system (the Sale Calendar): build models from verified sale history, backtest and calibrate forecasts, pool across sparse series, define abstention and scoring policies, and convert predictive distributions into calibrated consumer-facing verdicts. Ship production models and govern uncertainty communication; work with product/design and use AI agents/LLMs as part of the pipeline.

Design and own the verification engine that automates product truth verdicts: build confidence-scoring pipelines, adversarially robust models, a human-graded golden exam, and arbitration rulebooks. Raise machine-settled verdicts from 40% to 80% while maintaining 95%+ accuracy, embed eval-driven development, and scale solutions to millions of claims in production.

Own and evolve the production agent harness for the CEO office: design long-lived agent runs, verification systems, liveness checks, and token/model-routing. Build regression corpora, oracle-separated checkers, CI gates, and deterministic companions so automations never fail silently. Work directly with the founder, ship hardened systems, and measure falsifiable outcomes for a high-agency, production-critical AI platform.