Demos / Challenges / Hackathons
Talking about agents is useful. Building them is better. Three formats — a two-hour demo night, a six-week challenge, a two-day hackathon — all ending in something that runs.
Three formats
Pick the one that fits your appetite
Demo Night
Six to eight people, ten minutes each, everything live. No slides allowed — if it does not run, it does not count. The friendliest way to get your work in front of people.
Challenge
A real problem posed by a lab, a department, or an industry partner, with data or a simulator and a clear definition of success. Teams work on their own schedule and present at the next symposium.
Hackathon
Mixed teams formed on the spot, a theme announced at the start, and a demo to a panel at the end. Deliberately short, deliberately unpolished, deliberately fun.
Challenge lifecycle
How a challenge runs, start to finish
Challenges are the most substantial of the three formats and the one most likely to turn into a paper, a pilot, or a job offer. A good challenge has a real owner who wants the answer, data or a simulator that teams can actually use, and a success criterion you could argue about in a meeting.
- Teams of two to five, ideally spanning departments.
- Everything published at the end: code, prompts, evaluation, and what failed.
- No data leaves its owner without explicit permission and a written agreement.
- Costs are declared — a solution that needs a fortune in inference is a finding, not a win.
- Week 0Problem statement published
Context, data or simulator access, success criteria, constraints, and who to ask questions.
- Week 0Kick-off & team formation
A one-hour session: the problem owner presents, teams form, questions get answered in the open.
- Weeks 1–4Build
Teams work independently, with an optional weekly clinic for whoever is stuck.
- Week 5Submission
Working code, a short report, and honest numbers — including cost and failure cases.
- Week 6Judging & showcase
Live demos at the symposium, judged by the problem owner plus two community members.
- AfterWrite-up
Results and lessons go on the radar, and promizing work gets matched to a group or a supervisor.
Judging
What the panel actually rewards
Published in advance, applied to every submission. Note how little of it is about cleverness of method.
| Criterion | Weight | What the panel looks for |
|---|---|---|
| Does it work? | 30% | It runs on data the team did not choose, and the numbers are reproducible. |
| Evaluation quality | 25% | A sensible metric, a baseline to compare against, and honesty about failure modes. |
| Usefulness to the problem owner | 20% | Would they actually use this, or a version of it, next month? |
| Cost & practicality | 15% | Runtime, inference cost, and what it would take to keep it running. |
| Communication | 10% | Ten minutes that a non-specialist can follow. |
Calendar
Upcoming and past
| Event | Format | Dates | Problem owner | Status |
|---|---|---|---|---|
| Agentic Engineering Bootcamp | Pitch Competition | First week of Novemeber, 2026 | CCM | Open for teams |
Explore. Share. Connect. Build.
Come build something
Demo nights need presenters, challenges need teams, and hackathons need people who have never done one.