32dots HEIDELBERG AI
Session 13

Demo day

hard
USE 0 - 15 min

Watch the first demo — then critique it

  1. 1 Watch the first presentation (about 5 minutes).
  2. 2 Write down one question about: a failure mode, a data assumption, a privacy risk, or a scalability limit.
  3. 3 When Q&A opens, ask your question.
  4. 4 Listen to the answer. Was it satisfying? Did the presenter know or guess?
  5. 5 After Q&A, write a two-sentence private assessment: what was strong, what was uncertain.
  6. 6 Repeat for each team.

You asked at least one substantive question during the session and wrote assessments for each team.

UNDERSTAND 15 - 60 min

What makes a defensible architecture

🎯Problem framingWHY THIS PROBLEM🏗️ArchitecturerationaleWHY THESE CHOICES⚠️Failure honestyWHAT DOES NOT WORK
Key concept

A demo proves that a system can work once under ideal conditions. A defensible architecture demonstrates that the team understands why it works, where it breaks, and what it would take to make it reliable. The goal is not to impress — it is to be honest.

  1. ?Can you explain your architecture to someone who has never heard of n8n or Dify?
  2. ?What is the most fragile part of your pipeline — the part most likely to break in production?
  3. ?If your project continues beyond this course, what is the first thing you would change?
BUILD 60 - 90 min

Present and defend

Present your project with a live demo, then defend the architectural decisions in Q&A.

  1. 1 Open your system before your presentation slot. Test it once with your best test case.
  2. 2 Minute 0-2: explain the problem — who has it, why it matters, what they do without your tool.
  3. 3 Minute 2-4: run the demo — one real input, show the output, explain one architectural decision along the way.
  4. 4 Minute 4-5: show your test log — name one failure and what you learned from it.
  5. 5 Minute 5-10: Q&A. Answer what you know. Say 'I don't know' when you don't. Explain what you would investigate.
  6. 6 After all presentations: write a one-paragraph reflection on what you would do differently with one more week.
Deliverable

Working system + test log + one-paragraph reflection.

You built and defended a real AI system. What is the gap between what you presented and what you would need to deploy this in a real research environment? Make a list — that list is your roadmap.