Show HN: AI Dojo – open-source LeetCode-style trainer for AI prompts (github.com)

AI Dojo is an open-source, LeetCode-style trainer for AI literacy and prompt engineering that bundles short, focused exercises (A/B prompt selection, SQL puzzles, OpenAPI/YAML spec validation, and chat-driven rubric tasks) with instant auto-grading and qualitative feedback. Delivered as a small Flask backend + vanilla JS frontend, tasks are defined in a single tasks.yaml; submissions and grades persist in browser localStorage (and optionally server-side). Auto-graded work (SQL executed against in-memory SQLite/pandas, YAML parsed/validated) runs without any LLM key, while chat/rubric feedback can optionally call OpenAI or Azure OpenAI Chat Completions for richer qualitative assessments. The repo ships a modern UI (chat, file preview, export), a code sandbox, and simple install/run steps; it’s MIT-licensed and designed for extensibility. For the AI/ML community this is a practical, low-friction way to practice prompt design, API spec validation, and LLM interaction patterns with reproducible, shareable tasks. Technical trade-offs are explicit: the stack favors simplicity over scalability, tasks are YAML-driven for easy authoring, and LLM-based grading is an optional convenience (useful but potentially circular). Because many exercises auto-grade locally, the tool supports offline-safe skill checks while allowing teams to layer on OpenAI/Azure for nuanced feedback — a handy sandbox for teaching, onboarding, and iterating on prompt/evaluation techniques.