Meet Mitski. Study with purpose.

How to use it

1. 1Open Today and check the plan — it is already sized to the minutes you set in your Study Profile.
2. 2Work through the activities in order; tick each one as you finish.
3. 3Watch the completion ring close; finishing the last item ends the day with a small celebration.
4. 4If your Weakness Ledger has entries, your two weakest concepts appear here — tap one to see the full ledger.

The tech behind it

The plan is assembled on your device from a pool of activity templates, filtered to fit your available minutes. No AI is involved in building the daily plan — it is deterministic, so the same profile always gets a fair, balanced day.

Limitations — read this

The pilot includes cardiovascular activities and 24 source-linked foundation questions across respiratory, renal, neuroscience, endocrine, digestive and immunity topics. These additions await independent human review; they are not full specialty or exam coverage.

Completion tracks what you ticked, not what you learned; the honest signal of learning lives in Recall and the Ledger.

Your data

Your plan progress and profile stay on this device in guest mode. Signing in syncs selected study progress so you can continue on another device. Nothing about your study content is sent to analytics.

How to use it

1. 1Open Recall and answer the cards that are due — say the answer before revealing it.
2. 2Rate yourself honestly: Again, Hard, Good or Easy. The rating decides when the card returns.
3. 3Use cloze and image-occlusion cards for anatomy — hide the labels, rebuild the map from memory.
4. 4Cards you created in Card Studio and decks accepted from Lecture Rescue appear in the same queue.

The tech behind it

Scheduling runs on FSRS, an open-source spaced-repetition algorithm that estimates memory stability and difficulty. Reviews are scheduled when a memory is likely to be forgotten; this is an estimate, not an exact prediction. Every surface on the site that schedules review uses this one engine — there is no second scheduler anywhere.

Limitations — read this

FSRS optimises timing, not truth: rating yourself Easy on a wrong understanding schedules the error further away. The confidence buttons exist to catch exactly that.

A new deck needs a few days of honest ratings before the intervals become personal.

Your data

Guest review history lives in this browser's storage. Signed in, your review records sync privately so intervals follow you across devices.

How to use it

1. 1Pick a case and a difficulty — the difficulty sets your time and investigation budget.
2. 2Interview the simulated patient in plain language; order investigations that would change your mind.
3. 3Commit to a diagnosis and a management plan BEFORE the reveal — commitment is where the learning happens.
4. 4Read the examiner's rubric: history, efficiency, accuracy and safety are scored separately.

The tech behind it

The patient is played by an AI model that is given only what a patient would know — their story and symptoms, never the diagnosis. It cannot reveal what it does not have. Grading combines a deterministic rubric with AI assessment when you are signed in.

Limitations — read this

The patient is fictional and the encounter is a simulation — it is practice for reasoning, not a clinical reference.

AI grading can be imperfect; the deterministic rubric is the floor, and everything carries the educational-prototype notice.

Guest mode has a small daily allowance of patient turns; signing in raises it.

Your data

AI requests use store:false, but OpenAI may retain abuse-monitoring logs containing request and response content for up to 30 days by default, with longer retention possible for legal or safety reasons. Meyshan saves signed-in case results, including the diagnosis you enter, score and feedback; simulation state may also be saved on your device and in private account sync. Conversation text is excluded from analytics.

How to use it

1. 1Open an organ — it loads as an instant image first; tap to bring the full 3D model to life.
2. 2Drag to rotate, scroll to zoom, tap a glowing dot to read about that structure.
3. 3Run the labelling quiz: locate each structure, then say how certain you were — a lucky tap is not spatial knowledge.
4. 4Anatomy words inside Lecture Rescue lessons are tappable and open the atlas at the right organ, with a one-tap way back.

The tech behind it

The models are real 3D meshes rendered in your browser with three.js — no plugins, no downloads beyond the model itself. The poster-first pattern keeps the page fast: zero 3D bytes load until you engage, and low-power phones stay on the still image by choice.

Limitations — read this

Models are teaching simplifications, not surgical references — small structures are exaggerated for legibility.

Older phones may cap resolution to protect the battery; the quiz works either way.

Your data

Quiz results and confidence tags feed your device's weak-signal ledger; missed structures schedule into Recall. The models themselves are static files — nothing about how you rotate them is recorded.

How to use it

1. 1Choose a spotter image — histology, gross specimens, instruments and more.
2. 2Step one: describe the finding in your own words. Step two: say why it matters clinically.
3. 3Read the two-part score, see which discriminating words you captured and which to repair.
4. 4Tag your confidence — a pass you guessed your way through stays in your weak signals.

The tech behind it

Your description is compared against expert keywords with an AI-assisted rubric: half the marks for the finding, half for its significance. Occlusion masks let you turn any image into recall practice.

Limitations — read this

Keyword matching rewards precise vocabulary — a right idea in vague words scores lower than it deserves; use the feedback to sharpen terms, not to argue with the meter.

Uploaded images must carry no patient identifiers, ever.

Your data

Scores and confidence tags stay on-device for guests and sync when signed in. Adding your own image requires sign-in: it is stored in your private Spotter library until you delete it and is never added to anyone else's library.

How to use it

1. 1Work down the checklist — each item is a real step in the ECFMG/USMLE process with its documents and fees.
2. 2Tick what you have completed; the journey view keeps your place.
3. 3Use the linked official sources for anything involving money or deadlines.

The tech behind it

A hand-maintained checklist — deliberately no AI. Process facts change on official sites first, so every step links out to the source.

Limitations — read this

Registration windows, fees and rules change without notice. ALWAYS confirm on the official ECFMG/USMLE sites before paying or booking anything — this checklist is a map, not the territory.

Your data

Your ticks stay on this device in guest mode and sync privately when signed in. Nothing here asks for your official IDs or documents.

How to use it

1. 1Create cards by hand, import decks (csv, tsv or Anki .apkg), or accept a generated "Rescue: topic" deck.
2. 2Edit anything before it enters rotation — fronts, backs, tags.
3. 3Export to csv whenever you want your cards elsewhere; they are yours.

The tech behind it

One card model feeds one FSRS scheduler. Imported Anki decks keep their review history where the format allows, so your intervals survive the move.

Limitations — read this

Complex Anki note types flatten to front/back on import — check a sample before importing a huge deck.

Generated rescue decks remain unreviewed AI content unless an applicable independent review is recorded for that exact content. Check the source label and original reference before accepting a card.

Your data

Cards sync for signed-in students and stay local for guests. Export works either way — there is no lock-in by design.

How to use it

1. 1Paste the section that lost you, pick a topic, photograph the page or board, upload a PDF chapter, or just say it aloud — everything lands in the same editable text box first.
2. 2Answer 5–8 diagnostic questions and say how sure you were each time.
3. 3Read your Confusion Map: green is held, amber is shaky, red is broken or blocking. Start at the deepest red.
4. 4Pick a time budget — Emergency 5, Understand 15, Master 45 or Exam tomorrow — and repair, one concept per lesson.
5. 5Explain it back to an examiner or a patient, then accept the recall deck the session built for you.

The tech behind it

An AI model reads your material section by section and extracts a concept graph — concepts and the prerequisite links between them, never a summary. Source Truth chips grade every claim: directly supported by your material, simplified by Meyshan, added context, or needs review. The time budget changes the structure of lessons, not their word count.

Limitations — read this

Generated teaching is unreviewed AI content. Source chips describe provenance, not a clinical approval or a guarantee of accuracy. Verify claims against your textbook before using them.

A pasted section needs some mechanism in it; a bare list of headings cannot yield a prerequisite chain.

Photos and PDFs go through an AI reader: photos under 6 MB, PDFs under 15 MB. A chapter longer than the 24,000-character box comes back as its FIRST part with an honest note — trim to the section you need, or rescue the rest in another session.

Guests get one map a day with enough allowance to repair it; signing in raises the ceiling.

Your data

Meyshan does not write the original pasted lecture text to its application database. The generated map, questions, answers and lessons are saved on this device; cards you explicitly accept can also be saved to your signed-in Recall account. Meyshan records AI usage counters. OpenAI may retain request and response content in abuse-monitoring logs for up to 30 days by default, with longer retention possible for legal or safety reasons, even with store:false. Pasted text is excluded from analytics and shared libraries. Do not include patient identifiers.

How to use it

1. 1Pick the topic and the depth — Build it from the ground, Make it clinical, or Sharpen for the exam.
2. 2Mitski teaches the prerequisite chain bottom-up, one small concept per turn.
3. 3Every turn ends in a check you answer with stated confidence — a lucky guess is not knowledge.
4. 4If a turn doesn't land, ask for a simpler analogy, a clinical example, or Hindi.
5. 5The class closes with questions drawn from the same chain, and the session can be kept as a recall deck on your shelf.

The tech behind it

The Classroom is Lecture Rescue's engine pointed forwards: the same topic-graph extraction decides the chain, the same turn-based lessons teach it, and the same re-explanation styles adapt it. There is no second AI architecture — only a different entry: Rescue starts from your confusion, the Classroom starts from your curiosity.

Limitations — read this

Everything the Classroom teaches is AI output that Meyshan has not yet reviewed — the label stays on every screen. Verify against your textbook before the exam.

It shares the same daily study-AI allowance as Lecture Rescue: a deep class costs several lessons' worth.

Voice teaching is not part of this version.

Your data

Your check answers and weak signals stay on this device, with private account sync where supported. Meyshan records AI usage counters. Teaching requests use store:false, but OpenAI may retain request and response content in abuse-monitoring logs for up to 30 days by default, with longer retention possible for legal or safety reasons. Analytics excludes lesson text and the content of your answers.

How to use it

1. 1Write a promise in your own words — “repair cardiac murmurs before Sunday” — and pick its category.
2. 2Optionally link the module that will do the work and a target date, then post it to your wall.
3. 3Pin the promise that owns today; it appears at the top of Today.
4. 4Mark it kept only when it truly is, then add one honest line of reflection.
5. 5Archive what no longer matters; restore it if it starts mattering again.

The tech behind it

The wall is deliberately simple technology: no AI anywhere in it. Completion is never inferred from your activity — the KEPT stamp is always your own hand, which is the point. The only metric is commitments kept.

Limitations — read this

It cannot verify a promise was really kept — it trusts you, because a wall you can cheat is a wall not worth having.

This version is private-only: there is no sharing, no Study Circle wall, and no leaderboard — none is pretended.

Your data

Commitments are private by design: stored on this device as a guest, synced privately when signed in. The text of a commitment or reflection never leaves for analytics — events carry category names only.

How to use it

1. 1Pick your exam lens and system, answer, and tag how sure you were — confidence is the point.
2. 2Read the full explanation: why the right answer is right, and why EACH wrong option is wrong, with the real distinction.
3. 3Where it helps, an option also tells you what change in the stem would have made it correct.
4. 4A missed concept offers one-tap repair in Lecture Rescue; weak and guessed questions collect in your weak queue.
5. 5Spotted a problem with a question? Report it in two taps — every report reaches the review desk.

The tech behind it

Questions include a stem, options, teaching explanations, concept links, source basis and review status. Automated quality checks and an owner publication decision are not independent clinical review. Use the item-specific status; new pilot content may remain review-pending.

Limitations — read this

The bank favours fewer excellent questions over volume — coverage grows through the quality pipeline, not bulk generation.

Explanations carry an academic-review-pending label until each item is individually reviewed.

Your data

Your answers, confidence tags and weak queue stay on this device as a guest and sync privately when signed in. A problem report sends the question id, the category you picked, and your optional one-line note to the review desk — analytics sees the category only.

How to use it

1. 1Open the ledger to see recorded weak signals grouped by body system. Misconceptions appear before other signals, with recent entries first.
2. 2Choose an entry to bring its topic into Lecture Rescue. Check that topic before starting the repair.

The tech behind it

The ledger reads existing signals on this device and orders them with fixed rules. It does not call an AI model to identify or rank weaknesses.

Limitations — read this

An empty ledger means no readable signals were found here; it does not prove that you have mastered a topic.

The ledger only sees weak signals saved in this browser, so treat it as a pointer to what to review next, not a full picture of what you know.

Your data

Weak signals stay in this browser. When you choose a repair, the topic is handed to Lecture Rescue on this device, and Rescue clears the handoff once it has read it.

Analytics records how many entries the ledger shows. Opening a repair also sends the topic name and which module flagged it — never your answers or notes.

How to use it

1. 1Try the authored demo without signing in, or sign in to generate a paper. Choose your subject, exam track, mode, difficulty, question count and marks; an optional PDF can supply the source material.
2. 2Answer the paper, then finish to see scores and feedback. The demo is marked locally. If AI grading fails, the result explicitly uses a local rubric estimate.

The tech behind it

Generating a paper sends your settings and any PDF you attach to OpenAI; grading sends the questions, the marking guidance and your answers.

Generating papers needs sign-in. You get 12 units a day: a paper costs 2, grading costs 1, and failed requests are refunded. A site-wide AI limit, shared with other AI features (16 requests per 10 minutes, 80 per 24 hours), can pause you before your units run out.

Limitations — read this

AI questions, citations and scores can be wrong. Compare them with the original source and your course requirements; a mock score is not an official exam prediction.

Paper generation accepts 5–25 questions and up to 100 marks, with at least one mark per question. The optional source must be a PDF no larger than 15 MB. Service availability and shared limits can stop a request before the daily exam allowance is exhausted.

Your data

Each finished attempt saves a summary — when you finished, the subject, track, mode, scores and source — not your full answers or the PDF. It's kept in this browser and, when you're signed in, synced privately to your account.

Analytics only records that you started signing in from the Mock Lab, and whether saves succeed — never the paper, PDF or answers. Requests use store:false, but OpenAI may retain abuse-monitoring logs containing request and response content for up to 30 days by default, with longer retention possible for legal or safety reasons.

How to use it

1. 1Check the current Monday-to-Sunday board. Set your study profile, then generate recommendations using your exam track, available minutes and current subjects.
2. 2Add your own sessions and mark work complete as you do it. Regenerating keeps manual and completed sessions; previous weeks remain available.

The tech behind it

A rule-based planner combines the study profile with available Recall work and, when signed in, recorded weak questions. It does not send the plan to an AI model to generate the week.

Limitations — read this

Recommendations cover the available study material, not your entire curriculum. Check the board against your timetable and exam syllabus.

Completion is your own checkmark, not proof of learning. If a signed-in save fails or conflicts with another device, editing pauses so you can back up the visible draft and reconcile it.

Your data

Your boards are saved in this browser, one per week. Signed in, they sync privately to your account, and conflicting edits from another device are caught rather than overwritten.

Generating recommendations sends analytics the plan type, daily minutes and whether you are signed in; rebalancing also includes the exam track. The plan_created event does not include your session titles or written study content.

How to use it

1. 1Open Mastery Map ("Progress" in the menu) to see your recorded reviews, how many different cards you've practised, the cards you last rated Again or Hard, and your saved review schedule.
2. 2Use the links to continue Recall, bring a confusing concept to Rescue or begin a topic in Classroom. If history cannot be loaded, retry instead of treating missing figures as zero.

The tech behind it

The view calculates summaries from existing Recall records. It does not use AI to score your knowledge or write a new assessment of you.

Limitations — read this

Card ratings are self-reports. Review counts and due dates do not measure exam readiness, curriculum coverage or clinical competence.

As a guest, the history is whatever this browser has stored — including earlier guest sessions on the same device. Missing figures are labelled, never filled in.

Your data

As a guest, it reads the Recall history saved in this browser; signed in, it reads your privately synced Recall history.

This view reads the records without changing them and has no module-specific analytics event for the displayed counts. Normal page and navigation telemetry still applies.

How to use it

1. 1Enter a topic and choose all publications, reviews or randomised trials. Inspect the returned references, their source and retrieval time, then open the original PubMed record.
2. 2Stage a useful reference for the Future research workspace. Staging a reference is a handoff; you still choose whether to add it to a research project.

The tech behind it

Search looks up PubMed through the US National Library of Medicine's service (NCBI) and returns up to four article summaries. No AI is involved, so nothing is written for you. Only your topic (up to 120 characters) and the filter you chose are sent.

Search works without an account. Everyone shares one cache and one rate limit with PubMed; there's no personal daily allowance.

Limitations — read this

References are discovery metadata, not clinical approval or a systematic review. Read and appraise the original paper before relying on a claim.

Results can be cached or stale during an outage; the UI labels that state and the retrieval time. An unavailable or empty search does not establish that no evidence exists.

Your data

To stay fast and within PubMed's limits, results are cached under a scrambled (hashed) version of the search — never the raw topic or who searched.

References you stage for the research workspace stay in this browser until you save or dismiss them. Analytics records whether a search worked and how many results came back, and, when you stage a reference, where it came from — never your search or the reference itself.

How to use it

1. 1Record a check-in, reflect on your direction and keep track of tasks or academic work. Your ratings describe your own reflection; they are not a clinical assessment.
2. 2Type a question to Mitski without signing in. Live voice requires sign-in and microphone permission; use End to stop the conversation.

The tech behind it

Typed help sends your question, the mode you picked and up to four recent conversation messages to OpenAI. Requests use store:false, but OpenAI may still keep abuse-monitoring logs (see Your data).

Live voice connects you to OpenAI's realtime voice service. To set it up, we send Mitski's instructions, your name if we have it, and a scrambled (hashed) version of your email that OpenAI uses for abuse prevention — never the email itself. The store:false setting above covers typed help only.

Guests and signed-in students share the site-wide AI limit — 16 requests per 10 minutes and 80 per 24 hours — which other AI features also use. Guests are counted by a scrambled version of their network address that changes daily; signed-in students by their account.

Limitations — read this

Mitski can misunderstand a question or produce an incorrect explanation. She is a study and reflection companion, not a clinician or an emergency service.

As a guest, workspace changes last only until you leave or reload the page. Signed in, if a save fails or clashes with another device, your draft stays on screen and saving pauses until you download a backup and sort it out.

Your data

Signed in, your check-ins, reflections, tasks and academic work sync privately to your account. As a guest, they and your conversation exist only on the open page — nothing is saved to this browser — but you can download a backup.

Analytics never receives your reflections or conversations; only general navigation and save-status events are recorded. Typed questions and live audio go to OpenAI when you use them, and OpenAI may retain abuse-monitoring logs containing request and response content for up to 30 days by default, with longer retention possible for legal or safety reasons.