Learn what matters to you
A personal AI learning system.
We’re building connected teaching experiences and a research program to improve them. First comes Educator 1: iPhone lessons, practice and review.
Request testing accessTestFlight build for invited testing. Initial US invitations are planned for adults 18+; testing is free.
Our research approachSwipe from asking in Home to learning in Paths.
Illustrated product journey.
Home illustration: ask why astronauts float. Merak offers Gravity and orbit, with the description Explain why astronauts float and why the space station stays in orbit, and Build this path.
Separate Paths illustration, after the path is available: Gravity and orbit. The eight example lessons are Why objects fall, Mass and weight, Objects in free fall, Falling with the cabin, Moving sideways, Falling around Earth, Standing on the Moon and Explain weightlessness. Why objects fall is Next up; the illustrated path is at its beginning. The app has creation status and a ready preview between these selected moments.
01 · Educator 1
A clear sequence, with practice and review.
Our first teaching model plans eight connected lessons. Choose a suitable existing path or confirm a new request, then work through explanations, practice and later review.
- The ask
Choose a suitable path or confirm a new request. For a new path, Merak uses AI to plan lessons, write explanations and practice, and run automated checks. Follow its preparation status until it is available.
- The finish
Complete the lessons. Completed practice stays in your record; later recall is a separate check.
- Time to review
Merak schedules questions for another attempt, so you can check what you recall.
- A deeper pass
Further passes revisit the lessons with questions intended to demand deeper reasoning.
If you opt in, a notification can let you know when review is ready.
On the roadmapShared paths. We plan free access to eligible shared paths. Pro, our planned paid subscription, would help fund AI-generated paths that, when eligible for sharing, can serve more learners.
Four moves. One idea at a time.
Illustrative lesson: work through an explanation, answer before feedback and explain the idea in your own words.
A short card: one idea, your full attention.
Choose an answer and examine the reasoning.
Explain the idea, rather than only recognizing an answer.
Compare your reasoning with the explanation.
Back in Merak · Review
Come back to the questions.
Earlier questions return in the Review tab. Try answering again, then use the explanation to check your reasoning.
Your queue gathers what’s due. Start a review when you’re ready.
Swipe between the queue and a review question.
Illustrated Review flow. TestFlight build for invited testing.
Review-tab illustration. Three example questions are due: why older beans can produce less crema, what closes a habit loop, and what creates crema on an espresso. Start review opens the original question. In the illustrated review session, the learner chooses “The gas escapes over time,” then sees the explanation and Continue. This illustrates the flow, not a measured learning result.
On the roadmap
02 · Merak Lab
Understand learning.
Improve teaching.
Merak Lab is our planned program for testing how well Merak teaches, starting with Educator 1. The aim is to improve later recall and application, then extend what we learn to other teaching models and supported interfaces.
Practice records include attempts, answers and feedback. Relevant personal learning history would help us identify questions to investigate. Separate learning checks would test what people can recall and apply without help.
How the planned program would work
- 01
Teach and observe
We would use practice records and learner feedback approved for the research purpose to identify where teaching could improve.
- 02
Measure learning
We would test later recall and unfamiliar tasks without help. Completion and activity describe use; learning needs a separate check.
- 03
Test a change
We would compare a revised teaching approach with the current version and capable alternatives under consistent conditions.
- 04
Release and monitor
Learning evidence, content quality, reliability and full service cost would inform teaching-version releases. Monitoring a revision would guide the next test.
What should earn the next teaching versionEvidence of better later recall and application, within agreed limits for content quality, reliability and full service cost.
We plan to start with a learning researcher, outside methods review, engineers and designers.
Planned research, starting with Educator 1. Learning benefits remain to be demonstrated. Paths would be generated and checked automatically; separate studies would assess samples of learning experiences. Before a study, we would explain its data use and provide the required notice and choices. See our privacy policy.
03 · Teaching models
More ways to work through an idea.
A Merak education model is a teaching system: how it organizes explanations, practice, feedback and review. Foundation AI models provide language and reasoning.
Educator 1 is our starting point for structured paths. We’re developing Tutor 1 for conversational teaching, alongside visual and audio lessons. New versions are how we plan to improve each approach.
On the roadmap
04 · Learning assistants
One account. A conversation that keeps its context.
Planned assistants would carry relevant, permitted learning history between supported interfaces. Tutor 1 would guide the conversation; the interface could change.
A better next question starts with your last answer.
The assistant would draw on Corpus — your personal learning history: paths, answers and feedback. An earlier answer informs the next question; a fresh question checks your reasoning.
Gravity and orbit
“Objects fall together when air resistance is ignored.”
Your earlier written answer
The teachingTutor 1, then later versions, would guide questions, explanation and practice.
The interfacesMerak assistant, iMessage bot, WhatsApp bot and a talking avatar.
Your choicesChoose an avatar. Link supported surfaces to one account; control the history they can use.
Planned · scripted example
Today · you start the conversation
Press Play to see it in action.
MossA forest pebble spirit. Ideas take root.
Conversation · two synthetic voices in a scripted example.
Scripted roadmap previews, not available today. iMessage requires a supported Apple route. WhatsApp’s AI-provider rules restrict availability by service and country; Merak chat is our planned fallback. Avatar transcript.
Questions worth asking.
What is Merak building?
We’re building a personal AI learning system and a research program to improve how it teaches. Educator 1 on iPhone is the first experience. Our ambition is to connect different ways of learning through your personal history.
How does a question become a learning path?
Merak helps clarify what you want to understand and looks for a suitable existing path. If none fits, you can confirm a new request when creation is available to you. Educator 1 generates the lessons, explanations and practice, then runs automated content checks before a path becomes available. AI can still make mistakes.
How long is an Educator 1 path?
The current Educator 1 format plans eight connected lessons, with short explanations, practice and feedback. You work at your own pace across sessions. Further practice and the Review tab bring ideas back for another attempt; there’s no fixed time you have to finish in.
What do “education model” and “version” mean?
A Merak education model is a teaching system: how it organizes explanations, practice, feedback and review. Foundation AI models provide language and reasoning. Names identify teaching approaches; numbers identify versions. Educator 1 starts with structured paths. Later versions are intended to improve that approach through evaluation.
How will Merak get better at teaching?
Our planned research program would start with Educator 1. Practice records would suggest changes to test; separate checks would compare later recall and application without help. We plan to advance teaching versions with supporting learning evidence and defined quality, reliability and cost limits. Merak’s learning effects remain to be demonstrated.
Does completing a path mean I’ve learned it?
Completion records what you worked through. Explaining an idea without help, applying it in a new situation and recalling it later give you different ways to check your understanding. Practice and review support those attempts. AI explanations and feedback can make mistakes.
What does Merak remember about me?
Your personal learning history includes the paths you explore, your practice answers and feedback. It helps you return to your work. We plan to connect relevant, permitted history across supported Merak experiences, so future sessions can build on earlier ones.
Can the same path help more than one learner?
Yes. A shareable path that passes automated checks can serve other learners without being generated again. Your question and practice answers aren’t published with it. Reuse is how we plan to build a useful shared catalogue, while continuing to check and improve the teaching.
What comes after the first iPhone experience?
We’re developing conversational, visual and audio teaching. Our planned Merak assistant and talking avatar would build on relevant learning history. External messaging depends on platform rules; WhatsApp may be unavailable for our initial US use case. These experiences aren’t available yet.
Can I try Merak, and what will it cost?
Educator 1 is a TestFlight build for invited testing, with initial US invitations planned for adults aged 18+. Testing is free. We plan free access to eligible shared paths and a Pro subscription that includes new-path creation. Paid subscriptions aren’t available yet. You can request testing access.
No ads. We don’t sell your data. Read our research approach, privacy policy and terms.