AI · Drill
Free-response questions graded by Llama 3.3 70B via Groq, running entirely client-side. Each FRQ ships with a rubric — the AI grades against it, returns a 0–6 score with reasoning, and flags specific gaps for revision. Grading takes seconds, which is what makes it usable for actual study flow.
The rubric pattern is the key design choice: the AI isn't asked "is this good," it's asked "does this answer hit X, Y, Z criteria." Constrained grading produces consistent scores, which matters for treating practice as actual signal instead of vibes.