Governed AI: Power BI dashboards from a prompt or a template

Internship

• Field: DEV & AI
• Duration: 4 months
• Start: February 2027, or according to the university's calendar

Project

In a bank, a plausible wrong figure or a leaked row is worse than no answer. The AI only chooses from the project's approved vocabulary; code writes the SQL, the model, and the visuals, and every step is checked and measured.

Responsibilities

• Generate a governed Power BI semantic model from approved business names, declared joins, lists shown by name, and calculated fields as measures.
• Enforce the platform's row-level security (Chinese wall) in the model, outside the AI's reach.
• Generate dashboards from prompts by selecting measures, dimensions, filters, and visuals from a closed list; code writes the Power BI pages and visuals.
• Read an existing Power BI file or dashboard picture and rebuild it on the governed model.
• Apply guardrails: drop and count invented labels, refuse partial requests, name gaps in words, and ensure the same request always gives the same dashboard.
• Build a React product page with prompt, upload, preview, and download functionality according to an approved mockup.
• Create prompt and template evaluation sets, measure zero entitlement leaks and cost per dashboard, and compare governed versus free AI generation.
• Work using Scrum with two-week sprints, code reviews, and CI.
• Produce an end-of-study report and defence.

Candidate profile

• Final-year engineering student in software engineering, data, or AI.
• C# or another typed language and SQL.
• Interest in Business Intelligence (Power BI) and applied AI / LLMs.
• React / TypeScript is a plus.
• Rigorous, with an emphasis on tests, measurement, and clean design.