Open weights · runs on your laptop · no cloud
Drop in your PDFs — syllabus, slides, past papers. Notes vs. Me becomes the examiner: question after question, graded on the spot, with a heatmap of every topic you keep failing.
Six questions from a sample photosynthesis chapter. Graded instantly, tracked by topic — exactly how the real app works. Nothing to install.
Re-reading notes three times builds familiarity, not recall. Notes vs. Me replaces the third pass with the thing exams actually do.
Questions are weighted by your history: untouched material first, then whatever you keep getting wrong. Miss "Calvin cycle" twice and it quietly schedules itself for next round — the loop you just played above, running on your own notes.
Multiple choice is scored exactly. Written answers are judged by the model against a model answer — strict on understanding, generous on wording — with feedback that explains itself.
Parsing, question writing and grading all happen on your machine by default. No account, no upload, no telemetry. It works offline, because exam week doesn't always come with Wi-Fi.
A deliberately small pipeline — small is what survives exam week. One SQLite file holds all state; no vector database, no account.
Each PDF is parsed into paragraph-aware chunks of about 1200 characters.
Chunks are picked with weights from your attempt history — never-tested first, weakest next.
Gemma 3 (open weights, via Ollama) writes each question as strict JSON: topic, question, options, answer, explanation.
MCQs score deterministically; written answers are graded by the model, with feedback you can revise from.
Local by default. Gemma 3 runs through Ollama on your machine — about 1 GB of RAM, sized for the laptops students actually own. If a machine can't run a model, the app falls back to the same open weights served by Groq, and tells you plainly which mode you're in. The code is MIT-licensed; the models are open-weight; your notes never leave your disk unless you choose the fallback.