Academic Tutor

A personal learning system that turns PDFs into an active recall quiz, tracks your progress over time, and syncs cards to Anki.

Built on Make It Stick principles: retrieval practice, desirable difficulty, spaced repetition. No multiple choice — you write your answers from memory and Claude evaluates them.

Runs locally on your Mac using your Claude Code Pro plan. No separate API key required.

---

How it works

1. Ingest a PDF → Claude reads it and generates open-recall questions ranked by difficulty
2. Quiz → questions are served hardest-first based on your past performance; you type free-text answers; Claude scores them 0–10 with detailed feedback
3. Track → every answer is stored in a local SQLite database; your struggle score per question updates automatically
4. Sync → push cards to Anki via AnkiConnect with performance-based ease ratings

---

Setup

Requirements:

• Mac with Claude Code installed (Pro plan)
• Conda
• Anki + AnkiConnect add-on (optional — only needed for Anki sync)

Install:

conda create -n academic-tutor python=3.11 -y
conda activate academic-tutor
pip install -r requirements.txt

---

Usage

conda activate academic-tutor
cd ~/Desktop/Academic-tutor

Ingest a PDF

Drop PDFs in the pdfs/ folder, then:

python main.py ingest pdfs/yourfile.pdf

Generates 10 questions by default. Use -n to change that:

python main.py ingest pdfs/yourfile.pdf -n 15

Quiz

python main.py quiz

You'll see an interactive picker:

Which PDF do you want to study?

  0  All PDFs (23 questions)
  1  Can a biologist fix a radio?  (10 questions)
  2  Some other paper  (13 questions)

Enter number:

Type your answer freely and press Enter twice to submit. Claude scores it 0–10, tells you what you got right, what you missed, and why it matters.

Use --limit to cap the session length:

python main.py quiz --limit 10

Other commands

python main.py sources    # list all ingested PDFs
python main.py stats      # performance summary + weakest questions
python main.py sync       # push cards to Anki (Anki must be open)
python main.py reset      # wipe the database and start fresh

---

Progress tracking

Everything is stored in data/tutor.db (SQLite — stays local, never committed to git).

Questions are prioritised by struggle weight:

History	Weight
Never asked	1.1 — always near the top
Always wrong	1.0
50% correct	0.5
Always correct	0.1 — fades to the back

Run python main.py stats to see your average score and the questions you struggle with most.

---

Project structure

Academic-tutor/
├── main.py              CLI entry point
├── config.py            Settings (deck name, question count, etc.)
├── requirements.txt
├── pdfs/                Your PDFs go here (gitignored)
├── data/tutor.db        SQLite database (gitignored)
└── src/
    ├── claude_cli.py    Calls Claude via the claude CLI (no API key)
    ├── database.py      All DB reads and writes
    ├── ingestion.py     PDF text extraction + question generation
    ├── quiz.py          Interactive quiz loop + answer evaluation
    └── anki.py          AnkiConnect integration

---

Upgrading to the Anthropic API

Currently uses claude -p (Claude Code CLI) so no API key is needed. When you're ready to run it fully standalone, the only file to change is src/claude_cli.py — swap the subprocess calls for the Anthropic Python SDK.