MatrAIx began as a way to run an A/B test before any real user sees the change: a population of 8.3 billion persona agents tries each variant first, so teams learn what works, and for whom, before launch.
A new checkout flow, a pricing page, a chatbot's reply: product teams settle most of these decisions with an A/B test. MatrAIx began with one question. What if you could run that experiment before any real user sees it?
Running that experiment well needs three things: a realistic population of users, environments they can act in, and tasks that reflect real decisions. Those became the layers of the MatrAIx stack below. The same question, how does this work for this user, now drives our work on personal agent evaluation.
Collect structured and open-ended user feedback for market research, concept testing, and preference analysis.
Evaluate AI chatbots across task completion, customer satisfaction, helpfulness, safety, and multi-turn reliability.
Evaluate web prototypes and features across usability, presentation, navigation, latency sensitivity, and task completion.
Evaluate app features and workflows across functionality, responsiveness, task success, and user preference.
The persona population behind these studies now powers our work on personal agent evaluation, including the PersonalAgentEval benchmark.