01
Paste study text
The learner starts from a passage, note, or topic instead of a blank chat box.
Back to Projects
Education AI
Local-first Offline-ready Open Source Burmese-first
AI4Burmese Guided Tutor
A Burmese-first learning app that turns study text into step-by-step lessons, runs locally on the device, and keeps the experience simple for learners.
The goal is straightforward: give people a tutor that explains difficult topics in Burmese without needing a subscription, an API key, or constant internet access.
- Runs locally on a laptop or device.
- Begins with the learner's source text.
- Keeps the tutoring flow step by step.
- Uses an open-source implementation.
Personalized On-device LLM Open Source
The project page keeps the top image as a quick visual summary, while the technical note goes deeper into the build.
Why it matters
Education should not be gated by language, connectivity, or a paid account. This project is built around that idea: give learners a clear Burmese explanation that starts from their own material and moves at a manageable pace.
The result is a tutor that feels calm and direct. It does not try to be everything. It starts from a topic, stays with the topic, and gives the learner a step they can actually follow.
- It keeps the interaction Burmese-first.
- It removes the friction of subscriptions and API keys.
- It stays practical for offline and low-connectivity use.
- It is open source so the community can inspect and improve it.
Visual flow
How the tutor session moves
This is the basic lesson loop: the learner gives the tutor material, the tutor builds context, and the conversation moves one idea at a time.
02
Build lesson context
The tutor grounds the session in the source material and keeps the topic focused.
03
Teach one idea
The model explains a single point at a time so the lesson stays easy to follow.
04
Check understanding
The tutor waits for a learner reply before moving to the next step.
05
Continue the lesson
The flow loops until the learner has enough clarity to move on.
Video
See the tutor running locally
The video below shows the original working flow on the device. If you want the implementation details, the technical note explains the runtime and prompt structure.
The full technical write-up is available in the blog article.
How it works
Implementation in a paragraph
The implementation stays lightweight. Flask serves the app, a local OpenAI-compatible model runtime handles generation, and the browser receives streamed output so the lesson feels responsive.
The key design choice is the tutoring contract: keep the explanation close to the learner's material, move one idea at a time, and let the conversation continue only after the learner responds.
- Flask serves the web app.
- The local model runtime listens on 127.0.0.1:8000/v1.
- The browser UI runs on 127.0.0.1:5000.
- Responses stream so the lesson feels alive instead of stalled.
- The tutoring prompt keeps the model close to the source text.
Current status
What is already available
Laptop offline version
The local laptop version is already available and is the version this page points to most directly.
Instant web link
A simple web link is available so non-technical users can try the idea without setup.
Mobile version next
A mobile version is planned, with the same Burmese-first learning goal in mind.
- No subscription is required.
- No API key is required.
- Contributions are welcome from anyone who wants to help.
Source and related work
Where to go next
Source code
Local startup scripts, backend flow, and the implementation behind the tutor.
Read the blog article
A deeper technical note with flow, runtime details, and the lesson contract.
Padauk project page
The practical Burmese assistant layer in the same ecosystem.
Burmese GPT project page
The language foundation that supports the broader Burmese AI stack.