AITeachMe

Learning tools are good at storing material and increasingly good at answering isolated questions. The harder problem is continuity: turning a collection of slides, papers, and notes into an environment that remembers what has been taught, what remains weak, and what should happen next.

AITeachMe is a material-driven AI learning system built around that continuity. A course is not just a folder or a chat session; it is the boundary that connects source material, generated knowledge assets, teaching conversations, assessments, and an evolving learning profile.

The learning loop

The system organizes long-running work into five connected engines:

1. Ingest converts uploaded material into normalized Markdown and reusable assets.
2. Digest develops a learning plan, knowledge documents, and a knowledge graph.
3. Interact answers and teaches from course evidence and conversation context.
4. Examine generates questions, evaluates answers, and explains errors.
5. Profile records mastery, weak points, review tasks, and study plans for later sessions.

The value is in the reuse between stages. Parsed material should not be thrown away after one summary; an assessment result should affect later teaching; a generated document should remain traceable to the course material that supports it.

Workflow as a backend boundary

My work focused on backend architecture and implementation. FastAPI exposes the application boundary, while LangGraph workflows coordinate multi-stage AI operations. Repositories and shared infrastructure handle persistence, models, retrieval, storage, and observability without becoming the place where learning behavior is assembled.

That separation is particularly useful for AI products. A model call is easy to demonstrate; a workflow that can report progress, recover from partial failure, retain evidence, and produce durable state is what makes the feature usable beyond a demo.

Local first, with a path outward

AITeachMe can run with local SQLite and a filesystem-backed content store. PostgreSQL with pgvector and S3-compatible object storage provide a path for hosted deployments, but cloud infrastructure is not a prerequisite for the basic learning loop.

This is both a product and an architectural decision. Personal learning material can be sensitive, development should not require a full service stack, and a course package should remain portable. Optional parsing services, model providers, and hosted storage enhance the system rather than define its canonical data.

Current state

The project is moving from MVP toward early productization. The core course and workflow boundaries are in place, while deployment templates, public demonstrations, and community-facing polish are still developing. That status is intentional to state plainly: the strongest part of AITeachMe today is the coherent learning model and the backend needed to sustain it, not a claim that every classroom workflow is already complete.