A coding-practice platform that coaches you while you type, then makes you defend your answer, and never just hands you the solution.
Every AI coding tool has the same problem in education: it will happily give a student the answer. The student feels productive and learns nothing.
CodeLogs is built around a closed loop instead: recall, challenge, submit, defend, feedback, then a session tailored to what you got wrong. The AI is a behavioural coach and a constrained evaluator. It is deliberately not an answer generator.
Under the hood it's three independent services talking over Redis Streams, each owning its own data, with a dead-letter queue for anything that fails. I wrote a little over half of it.
Click any stage to read what it does, or press the button to walk the whole thing.
AI as a constrained evaluator, not an answer generator.
The architectural rule that shaped everything: no cross-service database writes. Each service owns its data and communicates only through events. It sounds like ceremony on a student project until the evaluation service falls over and you notice the sessions are all still intact.
The product rule mattered more. Deterministic checks run first and the model only ever layers structured feedback on top. A student can't extract the answer, because the system was never holding it.