You can't teach judgment with a slideshow. AI governance and compliance are full of grey-area decisions — the kind you only learn by making them and living with the consequences. A PDF of policies can't do that. A simulation can.
That's the idea behind GovernAI Studio — an AI-governance training simulator I built, where people learn compliance by working through realistic, persona-driven scenarios instead of reading about them.
Why training on judgment is hard
Most compliance training is a checkbox: watch the video, pass the quiz, forget it by Friday. It fails because governance isn't a set of facts to memorize — it's a practice of navigating ambiguity, weighing trade-offs, and defending a decision. You don't build that by consuming content; you build it by being put in the situation.
So the design goal was to put people in the situation — safely, repeatedly, with feedback.
Simulating the scenario
GovernAI Studio delivers a persona-driven interface for governance and compliance scenarios: you step into a role, face a situation, and make the calls, while the system plays the other parts around you. I owned the end-to-end frontend and extended the core simulator logic in a backend built on Django, Celery, and a hybrid RAG + LLM inference pipeline.
The pieces that make it convincing:
- RAG for grounding, not improvisation. The scenarios are grounded in real governance material through retrieval, so the simulation reasons from actual policy rather than an LLM's free-associating — which, in a compliance-training tool, is the entire difference between useful and dangerous.
- Personas that create real decisions. A persona-driven design turns abstract policy into a situation with a point of view, which is what forces genuine judgment instead of pattern-matching to a "right answer."
- Celery for the heavy lifting. Asynchronous processing keeps the experience responsive while the inference pipeline does its work in the background.
The bigger idea
The most interesting use of LLMs isn't answering questions — it's building environments people can learn inside. Grounded by retrieval so they stay honest, shaped by personas so they demand real thought, these simulations do something a document never can: they let you practice judgment before the stakes are real.
Building it convinced me that "AI for education" is less about generating content and more about generating situations. The full architecture is on the project page.
👉 See the simulator: www.divyakush.com/projects/governai-studio
Divyakush Punjabi — Full-Stack & AI Systems Engineer
🌐 https://www.divyakush.com · 💼 LinkedIn · 💻 GitHub
Top comments (0)