0
Due Today
0
Mastered
0
Learning
0
Unseen
0
Day Streak
0
Total Reviews

Weak Topics

Sorted by accuracy — weakest first. Answer questions to populate.

Topic Breakdown (All Questions)

Sources

📋 Content Pipeline

NBME forms tracked — Q = Questions PDF on disk  ·  A = Answer Key on disk

Form Q PDF A PDF Status MCQs Flash

Data

Surgery — NBME Practice

Questions from Surgery Self-Assessment forms

Exam Simulator

NBME-style timed blocks · 90 sec/question · blueprint-weighted sampling

Select Shelf
Block Size
🏥

Night Shift

You're the overnight resident on call.

Patients arrive with symptoms revealed one at a time.
Diagnose early for bonus points — but get it wrong and you lose HP.
Survive 10 floors. How far can you get?

🔍
Reveal clues
🎯
Diagnose early
= more points
❤️
Wrong = lose HP
⬆️
Pick perks
between floors
Difficulty
5 HP. Standard question pool.

Contents

Notes Question Generator BETA

AI-generated NBME-style MCQs from your FM notes — scored by a 6-dimension rubric

STEP 2 Study Site

An AI-augmented, single-file study platform for USMLE Step 2 CK shelf exams.

How It Works

This site is a self-contained study system built around two data sources: NBME practice forms (the question bank) and REMS shelf notes (the reference material). A content pipeline ingests PDFs, extracts questions via text parsing and OCR, assigns topics, and builds them into interactive quiz and flashcard engines. Everything runs in a single HTML file with no backend — your progress is stored locally in your browser.

Content Pipeline

NBME practice form PDFs are processed through a multi-step ingest pipeline:

1. Extract pdftotext pulls text from native PDFs. Image-based scans fall back to OCR (Tesseract). 7 subject-specific ingest scripts handle format variations across Psych, Surgery, OB/GYN, IM, Neuro, FM, and Peds forms.
2. Parse Questions are split into stem, options, correct answer, and explanation. Answer keys are matched to question PDFs. Each question gets a wrongExplanation per option — not just the correct answer.
3. Classify Keyword-based topic assignment maps each question to one of 145+ clinical topics. This powers the topic filter, weak-topic tracking, and heatmap diagrams.
4. Merge Deduplicated questions are merged into the MCQ and flashcard arrays. The Dashboard pipeline monitor tracks which forms are ingested, pending, or missing answer keys.

Features

Quiz

NBME-style multiple choice with full vignettes, 5 options, and per-option explanations. Filter by source or topic. Drill Missed replays only the questions you got wrong. SM-2 spaced repetition schedules reviews based on your performance.

One-Liners

Rapid-fire flashcard mode. Each card is a condensed clinical question — tap to reveal the answer. Same spaced repetition engine, same topic filtering. Built from the same question bank but formatted for speed review.

Web

Knowledge graph visualization. Topics are nodes, connections are shared questions. Clicking a node shows all questions that span that topic — revealing how concepts link across organ systems and specialties.

Exam

Subject-specific timed exam mode. Simulates a shelf exam by pulling questions from a single specialty, enforcing a time limit, and scoring at the end. Tracks your exam history.

Notes

REMS shelf notes for all 6 subjects (Surgery, FM, IM, OB/GYN, Peds, Psych) converted from PDF to structured HTML. Organized by organ system with collapsible sections, search, and 20 interactive SVG diagrams that glow based on your quiz performance via heatmap scoring.

Night Shift

A game. You're an intern on night shift — answer questions to survive floors. Wrong answers cost HP. Perks unlock as you progress: Overtime Mode, Attending Backup, Cram Session. Pulls randomly from the full question bank across all subjects.

QGen BETA

AI question generator. Reads the Notes HTML, generates new NBME-style MCQs using an autonomous loop inspired by AutoAgent. Each generated question is scored on 6 dimensions (clinical realism, medical accuracy, distractor quality, educational value, distinctiveness, format compliance) and only questions scoring 7.5+ are approved. The generation prompt self-improves across iterations.

QGen — How Questions Are Created

The question generator uses the same autonomous iteration loop as AutoAgent — a meta-agent framework where an AI iteratively improves its own output by measuring against a scoring rubric.

1. Select Pick the notes section with the fewest approved questions (ensures broad topic coverage across all systems).
2. Sample Pull 3 real NBME questions on that topic as style exemplars — the model learns the voice, difficulty, and vignette structure from actual exam questions.
3. Generate Claude generates 5 questions per batch from the notes content + exemplars. Output is strict JSON matching the app's MCQ schema (stem, 5 options, correct index, explanation, per-option wrongExplanations).
4. Score Each question is judged on 6 dimensions: format compliance (programmatic), clinical realism, medical accuracy, distractor quality, educational value, and distinctiveness (all LLM-judged). Weighted composite score determines pass/fail.
5. Evolve If scores stagnate, the system asks Claude to improve the generation prompt itself based on the specific weaknesses identified. The prompt evolves across iterations — better prompts produce better questions.
6. Approve Questions scoring 7.5/10+ are approved and appear in the QGen tab. You can browse them with full score breakdowns or practice them in quiz mode. Runs on your Claude Max subscription — $0 marginal cost.

By the Numbers

Built with Claude Code. Single-file architecture. No backend, no accounts, no tracking.
Your progress lives in localStorage — export it from the Dashboard.