AI-Powered Football Analysis Assistant

This project is a football analytics and betting assistant built with a FastAPI backend and a Next.js (React) frontend. The application:

• ▹ Fetches league, match, and team data from external football data APIs and stores it in PostgreSQL.
• ▹ Provides a Match Center where users can browse fixtures by league and season, and see detailed match views with xG, corners, cards, form, H2H, and odds.
• ▹ Computes model probabilities and edge / value-bet signals from prematch context and odds, and combines them with an AI-generated natural language analysis.
• ▹ Persists every AI recommendation in a prediction_records table, links it to the actual match result after full time, and exposes hit rate, ROI, and total profit/loss metrics for transparent performance tracking.
• ▹ Offers a Coupon Wizard that generates AI-assisted coupons based on selected leagues, date ranges, risk profile, and desired number of matches, using real market odds from the database.
• ▹ Emphasizes transparency with clear legal/ethical disclaimers on analysis and coupon pages, a performance summary block, and a “Why is this valuable?” section on the homepage.

Project Story and Challenges

The core goal of this project was to move from “one-off predictions” to a productized, auditable analysis system where value and reliability can be measured over time. Normalizing heterogeneous API responses across leagues and seasons into a single schema, correctly mapping odds to the right matches, and designing scheduler jobs that respect rate limits were early technical challenges.

On top of that, combining model predictions and edge calculations with LLM-based explanatory analysis required many iterations around performance and consistency. Instead of letting the LLM invent risk and confidence scores, we compute them rule-based in the backend and let the LLM focus on explaining those numbers, which makes the system’s behavior much more predictable.From a UX perspective, bringing single-match analysis, match center, team pages, and the coupon wizard together in one coherent Next.js interface meant designing cards and layouts that work both on mobile and desktop, favoring concise summary cards over raw tables where possible. Finally, logging every AI recommendation and resolving it with real match outcomes turned the project from a “nice demo” into a transparent, data-driven analysis tool where users (and we) can actually see how well the system performs over time.