What it is
KEIBA Insight does not claim to predict winners. It shows the evidence behind race-time projections so a racing decision can be inspected rather than trusted blindly.
The app compares every horse under today's exact conditions and makes each projected time traceable through source races, course and distance adjustments, confidence, and explicit data limitations.
How we built it
This is a real working application built on a complex Python codebase. During OpenAI Build Week, GPT-5.6 Sol helped us understand the codebase, find overlooked bugs and edge cases, and trace their impact across the prediction-support model, API, UI, and tests. Codex implemented the changes end to end, while I made the racing-domain decisions and validated them with automated tests and real race data.
The submission-period work includes robust fallback handling for sparse samples, clearer provenance for converted race times, and UI evidence that exposes confidence and data limitations instead of hiding them.
Why it matters
A prediction without reasons is difficult to audit. KEIBA Insight is designed for analysts and racing fans who want to see what supports a projection, what was adjusted, and where the data is weak.
Built With
- codex
- css
- duckdb
- flask
- gpt-5.6
- html
- javascript
- playwright
- python
Log in or sign up for Devpost to join the conversation.