Inside Chalks
Chalks is an education platform I’m building for teachers and coaching centers. Classes, assessments, question banks, payments, and student communication share one workspace.
Where AI fits #
In Chalks Proctor, Gemini interprets assessment results: where a class struggled, which wrong answers suggest a misconception, and what a teacher might revisit. Student analysis looks across attempts for patterns in performance.
The question bank uses AI for keyword suggestions and image transcription, making questions easier to organize and search even when the original material is an image.
The engineering underneath #
The application computes scores and statistics before constructing the model’s input. That gives the interpretation a concrete foundation: question-level correctness, skipped questions, score distributions, and frequently chosen wrong answers.
Access checks, rate limits, usage quotas, and cached insights are part of that workflow. Image transcription also checks that the question’s image hasn’t changed while the model was working before saving the result. These details matter as much as the prompt.
The stack includes Next.js, TypeScript, Convex, and Gemini, with authentication, payments, and live-class integrations around the core product.
