AI_Devs 4 is a Polish course on building applications with large language models. It runs five weeks: programming with LLMs, context management, nondeterminism and evaluation, working with AI, and generative applications. Every lesson ends with a practical task in capture-the-flag format.
My repository holds one solution script per lesson, some rewrites and side quests, and the biggest part: Dave.
Dave#
Dave is an autonomous agent that solves the course tasks. A planner breaks a task into steps. A model router picks a model for each step, either from OpenAI or a local one through Ollama or LM Studio. Generated code runs in a RestrictedPython sandbox. Tools cover HTTP, vision, audio, PDF, scraping, e-mail and shell, and Dave can hand work to sub-agents. It has an MCP server and a web interface.
Every LLM call is traced in LangFuse. Course data and model outputs are cached on disk, so runs repeat the same way. The coding agents that work in the repo have one rule: validate the answer before reporting success.
Stack: Python, uv, OpenAI API, Ollama, LangFuse, PromptFoo, Qdrant, MCP, FireCrawl, Docker.