How I Built the AI Academy
Designing an internal AI upskilling platform from curriculum to cohort cadence — the decisions that drove completion over content.
The standard approach to enterprise AI upskilling is to buy a licence to an online learning platform, push employees toward a catalogue of courses, and report completion rates to leadership. The completion rates are usually disappointing. The knowledge transfer is usually superficial. The behaviour change — people actually using AI differently in their daily work — rarely follows.
The AI Academy was built to do something different: deliver structured, cohort-based learning that combined curated curriculum, hands-on labs, and practical application challenges, with governance and tracking built in from the start. The goal was not course completions. It was measurable changes in how teams work.
Architecture
The platform is built on Next.js with a Supabase backend. Content is authored in MDX, which allows the curriculum team to write structured content with embedded interactive components — code examples, prompt templates, labs — without requiring developer involvement for every content update. OpenAI powers the interactive lab components: learners can experiment with prompts, see outputs, and iterate in a sandboxed environment inside the platform.
Supabase handles authentication, progress tracking, and the cohort management database. Each learner's progress is tracked at the module level. Cohort dashboards give facilitators visibility of where the group is and who is falling behind. Assessment results are stored per learner with timestamps, allowing the learning team to analyse which assessments are poorly calibrated and which topics have the lowest comprehension scores.
Curriculum Design Before Technology
The most consequential decisions in this build were curriculum decisions, not technology decisions. The learning arc was structured around three phases: awareness (what AI can and cannot do), application (how to use AI in specific role contexts), and creation (how to build AI-enabled workflows). Each phase had a fixed duration and a practical output requirement — learners had to demonstrate a concrete application of the skills, not just pass a quiz.
The cohort cadence — synchronous sessions every two weeks with asynchronous work in between — was the single biggest driver of completion. When the same curriculum was run as a self-paced programme, completion dropped to under 20%. Cohort-based delivery with a fixed cadence and peer accountability consistently delivered completion rates above 70%. The technology is largely incidental to this outcome; the pedagogy is the product.
Capabilities and Outcomes
The platform supports multiple concurrent cohorts with independent curricula, configurable assessment rubrics, and facilitator-controlled pacing. Content can be updated without a deployment cycle. The lab environment supports model selection so cohorts can work with different AI tools depending on what is approved in their organisational context. Completion tracking feeds a reporting dashboard that the programme team uses for executive updates. The primary outcome measure — change in self-reported AI use frequency, surveyed four weeks post-completion — showed a statistically significant increase in the pilot cohort compared to the control group that used the standard online learning approach.