Delivering AI and analytics products end to end for 5 enterprise clients in a 5-person squad (data scientists, engineers, product): problem definition, scoping, delivery and measurement. Ran product discovery for a national retail chain, defined 7 AI use-cases — all entered development — and put a working demo in the client's hands within 2 weeks. For a national logistics operator: demo in 3 days, then an image-analysis product covering 500 in-vehicle cameras continuously, replacing an entirely manual process. Led the model selection decision on a publicly funded domain-specific marketing LLM: evaluated 4 Qwen 3.5 variants on a purpose-built 500-question benchmark scored with an LLM-as-judge method; the 27B model did not win. Identified 12 metric inconsistencies in the evaluation pipeline and rebuilt it. Designed a 200,000+ record synthetic data pipeline; training on H200 GPUs.

Analytics and performance products for a portfolio of 10 global brands; progressed from specialist to senior over 5 years and took management responsibility in a 5-person team. Cut cost per install from TRY 100 to TRY 7 (93% reduction in 2 weeks) by running 4 audience segments against 5 creative x 5 copy variants in parallel, tracking at ad-set level and shifting budget to the winner. Mapped the customer journey end to end, built 6 behavioural segments at the points where users actually dropped out, each with an automated Mailchimp lifecycle journey — contributed to a run rate of up to 250 new customers per month on a single B2B brand. Built 10 automated Looker Studio reports and turned reporting into a self-serving capability used by 2 teams without going through me.