ReCall-nx

Review your distilled video course at n× speed, then test whether it stuck.

ReCall-nx is the front end over a .kb/ vector DB of distilled course notes (produced by a separate video-distilling pipeline). You watched ~160 hours of C and ~160 hours of C++ lectures once; ReCall-nx lets you re-review the distilled notes at whatever pace you choose, then quizzes your recall with answers cited back to the exact video + timestamp.

The n× idea

At 1× a review would take as long as the original course. At n× your review budget is original_hours / n:

Speed	Budget (160h course)	Depth	Shows
2×	80h	Deep	every concept, full notes + code
10× (default)	16h	Standard	important concepts, summary + code
20-40×	8h → 4h	Skim	key concepts, one line each
40×+	< 4h	Flash	headlines only

Speeding up raises the importance floor (low-priority concepts drop first) and shortens the per-card pacing timer, so the whole review fits the budget.

Sources

The source board lists every course in your KB. Click a card to toggle it into scope; review and quizzes span all selected sources at once (multi-select). "+ Add source" pulls in courses other people distilled; they appear as "distilling…" until their notes finish ingesting into the DB.

Three modes

• Review: set the speed and step (or auto-pace) through concepts at the chosen depth. Timestamps are clickable citations.
• Ask recall (you ask): ask questions answered only from your notes, with inline [collection/video @ hh:mm:ss] citations. Self-grade each answer (Nailed / Foggy / Blank).
• Quiz me (we ask): the app generates a question from a source in scope; you type an answer from memory and it grades you against the notes DB, showing which key points you hit or missed, a model answer with citation, and a running score. Toggle web-LLM cross-check to also credit correct answers phrased differently than the notes (stubbed in mock mode; wire to a real LLM-judge / web-search model with the backend).

Run it

npm install
npm run dev      # http://localhost:5173

Wiring the real backend

The UI currently runs on sample notes in src/data/mockData.js. To use the real .kb/ vector DB:

1. Stand up the FastAPI backend in backend/ (uv run uvicorn app:app --reload), which exposes /search and /ask over the distilled .kb/ vector DB.
2. Set USE_MOCK = false in src/lib/api.js. The /api/* calls are already proxied to http://localhost:8000 (see vite.config.js); fetchConcepts and askQuestion keep the same shapes, so no component changes are needed.

Structure

src/
  lib/nx.js          the n× speed model (budget, depth tiers, concept planning)
  lib/api.js         data layer; mock today, FastAPI tomorrow (same interface):
                     fetchConcepts / askQuestion / pickQuizConcept / gradeAnswer
  data/mockData.js   sample notes (SearchResult shape) + quiz questions/keyPoints
  data/sources.js    the source registry (on-board + community-available)
  components/
    SourceBoard       multi-select source cards + add-from-community
    NxControl         speed slider + budget/depth/concept stats
    ReviewMode        paced concept runner
    QAMode            recall Q&A (you ask) + self-grading
    QuizMode          active-recall quiz (we ask) + grading against the notes
    Citation          [collection/video @ ts] chips + answer renderer

License

MIT. Use it, fork it, bring your own KB.