Why I’m building it
Studying a technical subject involves more than getting an explanation. It means connecting ideas, working through problems and knowing which prerequisites still need attention. I’m developing Study Agent to bring those parts together, with mathematics as its first area of focus.
The goal is a learning companion that works with a learner’s own materials and maintains a structured view of their progress over time.
An explanation is useful. Evidence of understanding is what should shape the next step.
What I’m working on
My work spans the learning model, application architecture and implementation. The codebase includes material retrieval, persistent study conversations and workflows for reviewing proposed changes to the topic graph.
- Learning materials as context. Retrieval connects model responses with passages from imported materials, retaining source references that the learner can inspect.
- Structured learning state. Topics and prerequisite relationships provide a framework for planning. The design distinguishes encountering material from demonstrating understanding.
- Continuity across sessions. Conversations and application state are persisted so a study workflow can continue beyond a single model interaction.
Selected engineering decisions
Keep the learning record with the application. Python and SQLite support a local application with learner-owned data. The model’s conversation context is reconstructed from persisted records; it is not the sole source of truth for learning history.
Make model output reviewable. Graph changes and assessments need clear provenance and visible uncertainty. The project is designed around inspecting and correcting these decisions.
Use explicit model interfaces. Typed boundaries separate provider interactions from learning logic. Conversation summaries retain links to the messages they summarize, preserving a distinction between the original record and derived context.
Where it stands
Study Agent is an ongoing personal project. The architecture and learning workflows continue to evolve as I use and develop it. I’m focusing on making the full path from source materials to a useful study session coherent and inspectable.
It connects my professional interest in AI engineering with a personal interest in understanding difficult subjects deeply.