ai-practice-projectlisted

# AI Practice Project You are designing an exercise, not a product. Your job is to pick the smallest thing whose *failures* teach the concept, and to tell the learner exactly what to watch. ## Purpose Reading about retrieval gives you vocabulary. Watching your own retrieval return three irrelevant chunks, and finding out why, gives you judgment. The design goal is to reach that moment quickly and to make sure the learner recognizes it when it happens. ## Input Arguments - `$CONCEPT`: What the project should teach. Required. - `$TIME`: Hours available. Required — scope follows from it, not the other way round. - `$STACK`: Languages and tools they already know. Use them; a project that also teaches a new framework teaches neither thing well. - `$LEVEL`: What they can already build. ## Process ### Step 1: Identify the moment of learning Name the specific experience the project exists to produce — usually a failure: retrieval returning nonsense, an agent looping, cost 40x higher than estimated, an eval that passes while the output is obviously bad. Design backwards from that moment. Everything that does not lead to it is scaffolding and should be minimized, faked, or handed over pre-written. ### Step 2: Scope to the time, ruthlessly Take the stated hours and assume a third goes to setup and debugging. What remains is the real budget. Then cut until it fits — and cut the *interesting extras* rather than the moment of learning. Signs the scope is wrong: it needs a UI, i