More than a Hack · Barça Innovation Hub / 2026
IpurdIAr · Soccer GPT.
From raw football data to tactical conversations
At More than a Hack 2026, our five-person team built IpurdIAr, a conversational football analysis system. It connects positional tracking, events inferred from player and ball movements, and match commentary transcribed with Whisper so that a question can be answered using the relevant moments of a game. Our team won the Barça Innovation Hub challenge.
Whisper turboTracking dataEvent inferenceTactical featuresTool-using LLMMultimodal fusion
Making tracking data usable
The raw input describes the movement of players and the ball across a match. Our team summarized this high-frequency tracking stream into interpretable signals: pitch zones, player proximity, pressure proxies, defensive compactness and ball motion. We also inferred candidate passes and interceptions from the movement data instead of treating every event as ground truth.
Commentary meets structured events
We transcribed broadcast commentary using Whisper turbo and aligned relevant fragments with match moments. The question-answering agent could choose tools to query events, tracking-derived metrics or commentary, and use another source when information was unavailable. It was designed for post-match exploration, not a validated real-time scouting product.
Recognition and perspective
IpurdIAr won the More than a Hack 2026 award, and Barça Innovation Hub published an interview explaining the system. The value of the prototype was making multimodal match data explorable through natural language; the underlying code remains private and no independent performance benchmark is published.
Five-person hackathon project (Adrià Flores, Martina Vichino, Guillem Olivart, Roger Baiges and Hugo Recio). The source code is private. The organizer’s published interview documents the system and award, but no independent quantitative benchmark is available.