SSE Riga · One loop. Real users.
Build a useful AI service.
With Danielius.
In four weeks, launch something people use and make its everyday work run automatically.
For your SSE Riga course project. Teams form Thursday 15 October.
Your next step
Choose this challenge.
Introduce your team.
Teams form Thursday 15 October. Choose “One loop. Real users.” for your course project, then email Danielius. A rough idea is enough to join.
Start small. Make it useful.
A paid service, a useful free service or an internal tool can qualify. The users must be outside your team.
No idea yet? Pick a problem whose users you can reach this week.
Explore the 33 class ideas →An AI tool budget is available for up to three teams. Danielius confirms the support and terms with the selected teams.
The four-week plan
One useful service.
Better every week.
These are progress targets, not conditions for joining. Start with one result people want; then make the routine run without you.
- Week 1
Ship a first version
Practise daily with an AI coding agent. Put one useful result in front of 5 real users. A manual first test is fine.
- Week 2
Make it run daily
Automate the routine. Test its checks, failure handling and stop button. Reach 10 real users.
- Week 3
Get people returning
Aim for 30 real users and try for the first euro. Record repeat use, costs and your manual work.
- Week 4
Improve and prove it
Use feedback to test one change. Gather your numbers and rehearse the final demonstration.
Learn the model: delivery and improvement loops. See it in practice: my six builds. Then write your brief →
Final demo: at the end of the course project. SSE sets the date.
The final demonstration
Show that it works.
Your project must demonstrate these six things. Bring a working service and evidence of what happened.
- Real usePeople outside your team use it every week, or pay.
- An automatic routineA request or schedule starts the work. A checked result is delivered without you doing each step.
- A check that catches problemsShow one bad result your check held back.
- A safe response to failureShow it retry, stop or ask for help.
- Human controlShow results, costs, your manual minutes and a working stop button.
- One measured improvementUse real feedback. Compare the old and new version on the same saved examples.
Targets to aim for
30+ real users or 3 paying customers. At least two weeks of daily operation on its own. Evidence that people return week by week.
What counts as “running without you”?
The routine can receive a request, do the work, check it and deliver a result without you copying, clicking or sending each time. Record failures and prevent duplicate deliveries.
You still own the limits. Human approval for risky actions is fine: show where it happens and count the minutes. Inviting your first users by hand is fine too.
Label manual tests, recordings and made-up data clearly. Show what actually runs today.
See how to design the controls →How should we present the final demo?
Suggested 10 minutes: 2 for the user and problem, 4 for the loop running, 2 for a failure and the controls, 2 for the numbers and what you learned.
Keep a daily evidence log and use the weekly update template throughout the project.
Before you start
A few practical answers.
Do we need to know how to code?
You can start without coding experience. Use an AI coding agent, work in small steps and check each result. Build yours takes you from a rough idea to a first working loop.
What exactly is a loop?
A routine that runs again when a request or schedule triggers it: make something useful, check it, deliver it, then wait for the next run. It can use code, AI or both. A second loop uses feedback to test and improve the routine. Learn the loop model →
Does the service have to make money?
No. A useful free service or an internal tool also counts. Show real use by people outside your team. For a paid service, show how orders and payments fit the routine.
How does mentoring work?
Weekly mentoring with Danielius during the four-week challenge. Bring what you built, what users said and your biggest blocker. You can use the challenge to build your own project; working together afterwards is optional.
Which tools should we use?
Use tools that can run your routine reliably. The build guide covers choosing a tool, hosting the loop and testing it. Keep your first version small.
What about data and permissions?
Start with made-up data or data you have permission to use. Set allowed actions, recipients and spending limits. Require human approval for sensitive actions. Tell users what AI does and what data you keep.