I’m Dominika. I help teams ship AI that people actually use, with the risks known before launch, not after. I decide when it’s good enough and how people and AI share the work.
Product manager for five years. Building AI products since 2024.
I share product decisions, AI experiments and lessons from delivery.
From the first tests to a stable release: finding the real user problem, prototyping, testing variants, and bringing engineers, designers and domain experts into one plan.
Reviewers checking 200-page grant applications needed findings they could defend. The verifier compares the programme rules, the application and its annexes, and flags potential inconsistencies with links to the relevant sources.
Final verification went from 5 hours to 2–3, with a person still reviewing.
One wrong digit can mean the wrong order. We tested whether product names and catalogue numbers survived a translated call, and the findings shaped the glossary and the rules for alphanumeric strings.
Customers call in their own language; the agent switches on real-time translation for both sides.
In the pilot, most users chose the traditional route over the tool. I fixed the main campaign-creation flow first and shipped only what supported adoption, instead of working through a large backlog.
Quarter on quarter, in production: conversion of new users +36%, adoption +20%, monthly retention 62%.
Conversion+36%New users who reached their first campaign
Adoption+20%Users with access who have created at least one campaign
Retention62%Users who created a campaign and created one again the following month, in the second quarter
The three measures and how each one was defined. Growth is relative, quarter on quarter: two consecutive quarters in production. An active user is one who created a campaign.
Own experiment
The Last Mile, a browser game about shipping AI products. An AI tester won every run by taking shortcuts, so I rebuilt the scoring: a win now needs both launch performance and responsible decisions.
“She actively involved engineers in the product discovery process, valued technical input, and was always open to re-evaluating priorities when new information emerged.”
Define what an AI system can decide, who is responsible and what still needs work. Five questions, a worked example and a summary to share with your team. About 15 minutes.
