Why it exists
Reading unfamiliar source code is hard when you do not yet know the main abstractions or how they fit together. A useful explanation needs an order, not just a wall of generated text.
Portfolio / CASE STUDY
Public prototype
AI Tutorial Builder.
Turns a GitHub codebase into chapter-based tutorials, relationship diagrams, and approachable explanations.
MY ROLE
Repository analysis and learning experience
What I built
The application crawls a repository, filters files, identifies abstractions, analyzes their relationships, orders chapters, and generates Markdown explanations and Mermaid diagrams. An interactive file browser lets readers inspect the code alongside the generated material.
Project in view
Source for this capture
Engineering decisions
Give generation a structure
PocketFlow divides the process into explicit stages: fetching, identifying concepts, mapping relationships, ordering, writing, and combining. That makes the explanation workflow easier to reason about.
Reduce avoidable provider usage
LLM response caching and cache statistics support reuse of identical requests. File include/exclude patterns limit what the model needs to analyze.
Meet learners where they are
The README documents beginner-focused explanations, multilingual output, code walkthroughs, and diagrams rather than treating documentation as a source-code dump.
Capabilities in practice
- AI workflow decomposition
- Developer experience
- Technical communication
- Caching
- Web API integration
Status & scope
Public prototype. Generated explanations can be wrong or incomplete and should be checked against source code. Model access and repository permissions are required for the relevant operations. Related experiments include a portable Python CLI and adaptations to the upstream PocketFlow tutorial generator.