- 54%
- of support enquiries resolved by AI
- ~70%
- of marketplace cases covered after automation
- +15.85%
- checkout conversion, A/B validated
- 1 week to 4 hours
- monthly customer research analysis
- 23
- versions of a production AI prompt, each restorable
Where I’m most useful.
01 / AI in production
When AI needs to actually ship
Most AI features die between the demo and the release. The hard part is not the prototype, it is a system that survives latency, hallucination and unpredictable input while still feeling like one product.
02 / Delivery
When the team is faster than its process
Turning intent into a rigorous specification should not be the bottleneck. I run spec-driven delivery where AI does work at every step and a human decides at every step.
03 / Discovery at scale
When there's data but no clarity
Research that costs a week gets skipped. With thousands of touchpoints, manual synthesis is impossible, so I build pipelines that turn raw conversations into structured decisions.
Selected Work.
- 01Getting six product people to change how they work6 of 6
- 02From prototype to pull request, with AI at every step2 sessions
- 03Reading 1,182 customer conversations in a day1 wk to 4 h
- 04Making the pricing page make sense+15.85%
- 05Answering half the support queue without hiring54%
- 06A ship gate for LLM quality23 versions
- 07One inbox for 70 marketplace channels~70%