How openai/build-hours works
Project type and purpose
• This repo is a “monorepo” of small, self-contained demo applications (numbered 10–13, 2–9, etc.) showcasing different OpenAI API use-cases (e.g., distillation notebooks, recommendation engine, agentic tool calling, image generation, fine-tuning examples, real-time demos, etc.).
• Each numbered directory is effectively its own mini-project (some in Python, some in Next.js/React, some pure scripts/notebooks).
File structure
• Top-level: .gitignore, LICENSE, README.md, then dozens of numbered subdirectories (10-distillation, 11-recommendation, …, 13-image-gen, 2-assistants, …).
• Each subdirectory generally has its own source code, configuration, requirements/package files, and sometimes a docker-compose or notebook.
• Patterns:
• Python demos: app.py, helper_functions.py, config.py, Docker compose + Qdrant or SQLite.
• Next.js demos: imagegen-demo with src/app/api/* routes, React components, Tailwind.
• Jupyter notebooks for research purposes.
• Utilities and scripts for ingest/upload, evaluation, fine-tuning.
README
• The top-level README introduces “Build Hours”: monthly events for startups and developers to explore OpenAI’s API.
• No single architectural overview is provided—each subfolder is a standalone exercise.
Detail without overthinking
• At the monorepo level: group each numbered example as its own “module” node, showing that it has its own tech stack.
• At the demo level: drill into one representative folder and break it into UI, service, data, external API.
• Moving parts: local code, containers, scripts, external APIs.