AI/ML tools, pipelines and prompt engineering for practical handicapping
Key Questions
What is the main strategic focus of the AI/ML tools and prompt engineering work for handicapping?
The focus is on reproducible data ingestion using DuckDB and Parquet, turf-aware feature engineering, temporal validation for long-horizon targets, and prompt banks that convert model outputs into stake-sized wagers and dashboard explanations.
What deliverables are prioritized in this highlight?
Prioritized items include a prompt bank, two app-ready templates, and worked examples validated on sample racecards, with status currently marked as developing.
How might private or local AI tools apply to racing-specific setups?
Articles on tools like Jan, Msty, and AnythingLLM are noted as potentially adaptable for building private LLM setups tailored to racing data and workflows.
Strategic focus remains on reproducible ingestion (DuckDB/Parquet), turf-aware feature-engineering recipes, temporal/holdout validation for long-horizon targets, and actionable prompt banks that translate model outputs into stake-sized wagers and concise dashboard explanations. Deliverables prioritized: prompt bank, two app-ready templates, and worked examples validated on sample racecards. Additionally, an article on private/local AI tools (Jan, Msty, AnythingLLM) was noted as potentially adaptable for racing-specific private LLM setups.