A lot of AI work comes down to picking from a fixed list: routing tickets, flagging spam or fraud, qualifying leads, moderating comments, or detecting intent. Running a full LLM for a one-word answer gets expensive at scale. Decision models score every option in a single request and return an answer with a confidence score, faster and cheaper. The workshop explores working with real data: when to trust the cheap model, when to escalate to an LLM, and what each saved dollar costs in quality. No engineering background is needed. The agent writes the code; you decide how to frame the question for the model and assemble the pipeline. Work solo or team up with a neighbor. An introductory primer covers the basics, and organizers will help with setup. What to bring: - A laptop with a coding agent (Claude Code, Codex, Cursor, or similar) - A card for API calls if your subscription does not cover them (expect to spend a few dollars at most) Registration: via the link in the organizer's Instagram profile.