TR2 Retrospective

Diverse, cross-cultural teams helped students learn better.

The most important benefit students raised: working in groups mixed across countries broadened their perspective, gave them real exposure to different cultures, and made the team itself work more effectively — not despite the mix, because of it. That's why TR3 teams are deliberately built for country diversity rather than left to self-select.

The UI got built too late — a last-sprint scramble instead of steady progress.

Teams focused entirely on the pipeline and evidence until the final sprint, then tried to build a real interface for their work in the time left before Demo Day. The result looked exactly like what it was — a rushed afterthought, not a product. Progress was judged on GitHub commits alone; nobody was looking at whether the thing actually looked like something worth showing. This trimester, UI ownership is an explicit part of R6 starting Sprint 3, not a final-sprint scramble.

Initial resistance turned into real financial literacy — and it showed up in students' careers.

Many students pushed back early on having to learn market intelligence in an IT-discipline course. By the end, that resistance turned into genuine grasp of the material. Two students said this project came up in their Master's admission interviews — discussing market dynamics and how they learned to work with data — and it helped secure their place. Three more discussed the same project in strong job interviews, and two of those were accepted. What came up specifically: the ten-week Scrum rhythm, the discipline of pulling data together systematically week after week, and a genuinely deeper understanding of market dynamics than they expected to walk away with.

Deep, practical LLM literacy — the one course that actually taught agentic AI.

This was the one course that encouraged real, hands-on LLM use rather than treating it as something to work around. Students who went through the full ten weeks came out with a genuinely deeper understanding of how LLMs work and how to deploy them agentically — not just the free tiers of ChatGPT or Gemini directly, but broader access through sites like OpenRouter, which opens up a much wider range of models for free. By the end, students could speak fluently about the tricks the market actually uses to get real leverage out of LLMs, not just the basics.