Turning textbook pages into lessons that work offline.

Preparing a lesson starts with source material, but teaching requires more than a summary. Teachers need explanations, teaching notes, questions, activities, and homework they can use together. A cloud-dependent tool also introduces a practical constraint: a classroom may not have reliable internet.

ChalkLens brings those tasks into a desktop workflow. The product is built around a teacher’s own PDFs and scanned pages, with local inference rather than a remote model call.

I built the Flutter desktop application and its lesson workflow, using Dart, Riverpod, PDFKit, OCR, and local Gemma 4 inference through LiteRT and Google AI Edge. The output is a lesson kit containing explanations, teaching notes, quizzes, activities, and homework.

The application also includes a Present mode for teaching from a laptop and a Student Help flow grounded in the lesson context. That connects preparation to what happens during the class, instead of ending the experience at a generated document.

On-device inference makes the desktop runtime part of the product. Document extraction, scanned-page recognition, model execution, and the teaching interface all need to fit into one usable application. The workflow cannot assume that a hosted model will always be available.

Keeping student help tied to lesson context also gives the interaction a specific purpose: helping with the material the teacher is using. It is a narrower, more useful workflow than an open-ended chat window.

The result is an independent product build demonstrating document processing, local AI integration, and a complete desktop teaching workflow. The source repository is linked below so the implementation can be inspected.

This study describes the built workflow. Classroom adoption, learning outcomes, and time savings have not been measured here.