AI Documentation Generator

Built an AI documentation system using the Claude API with structured output parsing that auto-generates Markdown documentation across a multi-language codebase (Python, C++, JavaScript, Vue.js). Reduced a 3-day manual documentation process to 4 hours — producing consistent, structured technical docs for 34+ ROS packages and 150+ Vue components.

Designed a multi-language documentation pipeline that analyzes source files across the entire Productive Robotics codebase — C++ motor control (EtherCAT, CAN bus), Python ROS nodes (executive interpreter, welding drivers, trajectory planning), JavaScript/Vue.js (tablet UI components, ROS communication layer), and YAML configuration (robot models, recipes, launch files). Each language has a tailored analysis prompt that understands the framework conventions: ROS node structure for Python, class hierarchies for C++, component lifecycle for Vue.js.

Integrated with the Claude API using structured output parsing to generate consistent documentation in a standardized Markdown format: module overview, public API reference, key dependencies, configuration parameters, and usage examples. The structured output ensures every generated doc follows the same template regardless of the source language or module complexity. A post-processing step cross-references generated docs to add hyperlinks between related modules (e.g., a welding driver doc links to the executive interpreter step that calls it).

The system reduced documentation effort from a 3-day manual process to ~4 hours of review and refinement. Output is served via a Flask-based documentation server (pr-docs/server.py) that auto-renders Markdown with navigation, search, and cross-module linking. The doc server hosts ~200 files covering component docs, feature guides, deployment procedures, and knowledge transfer materials — all maintained as living documents that regenerate when the underlying code changes significantly.