AI Certification Copilot: giving certification study a direction

The starting point: deciding what to study next

When I began preparing for an AI certification, I wanted a clearer way to judge my next step. Finding more material was easy. Understanding which concepts I knew, which answers were lucky, and where to spend my next study session was harder. A total score alone did not explain those differences.

The project began as GAL Sprint, a diagnostic tool for my own learning needs. On July 20, 2026, it expanded into AI Certification Copilot, with the aim of connecting an initial diagnosis to study and review that I could return to over time.

The decisions: make the gaps understandable

I chose a sequence of profile, certification-path comparison, diagnosis, then study and review. The diagnosis considers correctness, domain weighting, confidence, and time. Together, these signals give a learner more context for deciding where to focus.

An important correction came from the limits of sampling. A diagnostic could miss areas of the official syllabus, even when the score looked strong. I added a full-coverage check to make that distinction visible. Doing well on sampled questions and checking every relevant topic are different kinds of evidence.

The implementation: carry the diagnosis into review

GAL now supports the complete study flow with 32 original scenario questions, coverage checks across 44 topics, review, mock exams, and wrong-answer review. Profiles, mock-exam records, and PMLE/PCA progress are designed to save to D1 after login. GAL diagnosis and topic coverage currently stay in browser storage, with summary and JSON exports available. Cross-device continuity and the shared-browser experience still need work.

I defined the learning problems, priorities, scope, and acceptance criteria, working with AI coding agents on implementation and review. The expansion keeps its limits explicit: PMLE and PCA each have a 12-question diagnostic beta. The nine-certification roadmap describes possible paths; it does not mean nine complete courses or question banks.

The evidence: a working tool and a clearer standard

The website is public, while access to private learning data requires login. The result is a usable study tool and a record of how my own preparation changed the product. I have no evidence of other learners’ pass rates, so that is not an outcome claimed here.

The strongest lesson was to turn readiness into questions I can check: which topics have I covered, which remain uncertain, and what should I revisit? Future additions can be evaluated against those same questions, with content quality and syllabus coverage validated as the scope grows.