PizzaTipz

A daily pipeline that pulls fixtures across eight sports, runs structured Perplexity research per fixture, has Claude synthesise constrained selections, then auto-settles results and feeds calibration back into the next day's prompt.

Stack

Engineering highlights

• Discover → research → analyse → build → settle → learn loop with a self-improving calibration step.
• Structured LLM orchestration: per-fixture research prompts, constrained JSON outputs, and confidence/odds bounds.
• Automated settlement and per-leg grading from a free results API, with a market-bandit + drift-detection learning layer.