For 5 years at American Express, I led product analytics — A/B/n testing across 6+ international markets, an ML anomaly detector for revenue-critical pages, SEO at the scale of 150K+ URLs, and a team of 6 data scientists.
In May 2025, I joined DataDrew as AI PM to build Drew AI — an LLM-powered analytics platform for Shopify D2C brands. We turned Meta Ads, Google Ads, and GA4 data into decision-ready insights, with confidence scoring and human-in-the-loop controls so founders could actually trust the output.
Since June 2026 I've been at BusinessNext, building AI-native CRM for banks. Different constraint set entirely: an answer isn't shippable because it's right, it's shippable because it passed role-based access, masking, domain routing, controlled output, and left an audit trail a regulator can read.
I write about AI products, evals, and the gap between what's hyped and what works. Often with metaphors. Sometimes with my 2-year-old daughter as the test subject for whether an idea actually makes sense.
AI-native CRM for banks. Agentic workflows, guardrails and governance, RBAC and masking, evaluation frameworks, human-in-the-loop controls for regulated deployments.
Built Drew AI — an LLM-powered analytics platform for Shopify D2C brands. Multi-agent reporting, eval infra, MCP integrations, GTM, pricing.
Led A/B/n testing across 6+ markets impacting $100M+ revenue. Built ML anomaly detection, ran SEO at 150K+ URLs, mentored a team of 6 data scientists. Drove cross-functional product initiatives including Open Banking integration and a Bot Identification System for cleaning experiment data.
Built income prediction model with linear regression. Owned model segmentation, uplift reporting, governance, and stability monitoring — my first taste of shipping a model into production.
Implemented Theory of Constraints — reduced inventory ~17%, improved OTIF rates. Marketing analytics for Wire Rod products: bucketing, raw material forecasting, 50K-ton production planning.
A depth map across the things I actually touch. Five squares per skill: five = expert, three = working knowledge, one = familiar.
I post regularly on LinkedIn — usually short essays with a story or analogy at the centre. Long-form essays will live here soon.
For now, the writing lives on LinkedIn. Topics I'm chewing on: agentic systems that actually work, why eval infrastructure beats prompt tweaking, MCP as the connective layer, and what e-commerce analytics looks like after dashboards.
Read on LinkedIn →