JADE LEE.
← ALL PROJECTS · /02 · AI FOOTBALL SCOUT

SIDE PROJECT · IN PROGRESS · JUN 2026 —

AI Football Scout.

An EPL scouting assistant that answers player-stat questions through an Orchestrator/Worker LLM pipeline, verifying every number against source data before it reaches the user.

01 · THE PROBLEM

Asking an LLM a player-stat question is easy; trusting the number it returns is not. Models hallucinate stats and reasoning that sound plausible but aren't backed by the source data.

02 · KEY FEATURES
01

Orchestrator/Worker pipeline

An Orchestrator routes stat questions to Worker LLM calls on Gemini 2.5 Flash, keeping each worker focused on a narrow, verifiable task.

02

Dual judge verification

Every answer passes through a deterministic numeric checker (pure code, zero LLM cost) and a structured-output LLM judge that catches unsupported numbers and reasoning, retrying failed answers automatically.

03 · WHAT I OWNED

Designed and built the Orchestrator/Worker pipeline solo, end to end.

Built the dual verification layer — a deterministic numeric checker plus a structured-output LLM judge.

04 · WHAT'S NEXT

Currently expanding coverage across EPL player stat questions and hardening the verification layer.