I build L&D programs, with certifications as a speciality.
These are my AI experiments.

I've run exam design workshops and psychometric reviews professionally at TikTok and Google. This pipeline uses 14 AI agents to replicate that process: evidence gathering, SME simulation, bias review, and blueprint generation, with those standards in mind and not as a certified claim. Five human-in-the-loop gates keep a person in the decision chain. I built it independently; it's now in use at Intuit.

Performance-based exams are the hardest part of certification design to get right. This system provisions an isolated coding environment per candidate, runs deterministic validation, then uses a four-dimension AI rubric (Delegation, Description, Discernment, Diligence) to assess how well they worked with AI. It grades what humans currently review, and it's now in development at Intuit. Its successor, the flagship above, moves the whole exam live and local.

Having designed learning programs at Google and TikTok, I wanted to see if AI could handle the full pipeline: audience pain discovery, core principles definition, Bloom-aligned curriculum design, formative assessment, and experience mapping. Eleven agents collaborate through eight human-in-the-loop gates, enforcing instructional design methodology at every step. Free learning funnels toward paid certifications. Built on the same orchestration pattern as the JTA Exam Builder.