Milestone map
Milestone map
2 milestones
Ship a real AI system to real users
6–10 weeks (15–20 hrs/week)
Build and deploy an AI system that solves a real problem. At least 3 people other than you must use it. Keep the code public, evaluate it properly and measure what it costs and how fast it is. This is what the Honors blind review reads.
Proof required
Link the live system, its public repo, an evaluation report (what you measured, on what data, with what results) and a cost and latency note (cost per request or per user, p50/p95 latency), plus proof that at least 3 other people used it.
What gets checked
- It is deployed and works for real users, not just in a demo.
- The evaluation report has real numbers on a stated test set.
- Cost and latency are measured, not guessed.
Common mistakes
- A demo nobody else used.
- An evaluation made of a few hand-picked examples.
Powstik Guide
A deployed AI system used by at least 3 other people, with a public repo, an evaluation report and a cost and latency note.
Steps
- Pick a problem people near you have.
- Ship the smallest useful version first.
- Build the evaluation set before you tune anything.
- Get at least 3 people to use it and fix what they hit.
- Measure cost and latency under real use.
- Write it up and submit.
Template
README Problem: Who uses it: Live link: How to run it: Evaluation (data, method, results): Cost and latency: What users changed:
What gets sent back
- No real users.
- No evaluation report.
- Cost and latency missing.
Resources
Foundationstart here
Depthgo deeper
What a verifier looks for
- Can you use it?
- Do the evaluation numbers hold up?
- Are cost and latency measured?
- Is the user proof real, and is it the person's own work?
You'll sign in first, then come straight back here.
Record a 10-minute walkthrough
1 week (5 hrs)
Record a 10-minute walkthrough of your system: the problem, a live demo, the code and how it fits together, the evaluation and what you would do next.
Proof required
Link the recorded walkthrough (about 10 minutes).
What gets checked
- It shows the system working live.
- It shows the code, not just the screens.
- It explains the evaluation results honestly.
Common mistakes
- Slides only, no live system.
- Hiding the parts that do not work.
Powstik Guide
A 10-minute recorded walkthrough of your AI system.
Steps
- Plan: problem, demo, code, evaluation, next steps.
- Practise once, timed.
- Record it and submit.
Template
Walkthrough plan Problem: Demo: Code tour: Evaluation: Next:
What gets sent back
- No live demo.
- No code shown.
Resources
What a verifier looks for
- Does the system in the video match the repo?
- Is the code shown and explained?
- Are the results presented honestly?
You'll sign in first, then come straight back here.