Inspiration
A coach can finish looking at a dashboard and still have the same question: what should we work on at training? MatchFlow connects match review to a practical session plan. It is designed for coaches who know football but do not want to interpret a wall of statistics before deciding what to do next.
What it does
Choose a country to become your demo team, then explore its matches. MatchFlow brings together a simple overview, pitch maps, recorded action sequences, and a conversational coaching guide. Ask “How can I improve our defending?” and follow up with the number of players and time available. The guide can explain a possible focus, link to relevant match examples and contrasting moments, and suggest an editable activity, coaching cue, and check for the next game.
The coach stays in control: inspect the moments, choose a focus, adapt the practice idea, and save or export the session brief. Detailed formulas, source information, and analytical limits have their own Data & methods page.
How we built it
The application uses JavaScript, HTML, CSS, a Node.js backend, and Vercel hosting. A reproducible pipeline compacts StatsBomb Open Data into records for all 115 matches from Euro 2024 and the 2022 World Cup. The app can also refresh selected match records from the public source.
Shared analysis code reconstructs possessions, measures progression and shot creation, and connects regains or losses to subsequent shots. Tournament comparisons exclude the selected team's observations. Replays show recorded actions on a pitch, including the opponent response after a loss; they are schematic event sequences, not match video or continuous tracking.
The optional OpenAI-powered guide receives evidence recomputed on the server, sourced coaching summaries, and recent conversation. It can answer follow-up questions naturally and adapt proposed practice activities to the coach's constraints. Its response schema limits citations to supplied match records and coaching sources. API credentials stay on the server.
Challenges we ran into
Our biggest challenge was translating analysis into a useful coaching conversation. Early versions overwhelmed the working screens with statistics and forced answers into a rigid report. An overly restrictive check also rejected legitimate numerical explanations. We simplified the interface, moved technical detail to Data & methods, preserved both sides of the conversation, and let the guide write a natural response with inspectable sources.
Another challenge was showing the whole relevant sequence. For defending after a loss, replaying only our team's possession hides what happened next. MatchFlow joins that loss to the opponent's response and offers a contrasting example for review.
Accomplishments
We built a review-to-session flow on public match records, with selectable demo teams, interactive pitch maps, paired examples, conversational help, and editable, exportable practice plans. The application includes automated checks for the data, analysis, evidence chains, conversation history, and guide endpoint. Synthetic live conversations have exercised follow-up explanations and constrained session planning.
What we learned
More visible statistics do not automatically make a better coaching tool. Coaches need a clear question, a relevant moment, and a practical next step. Useful AI guidance also needs to distinguish what the records show from general teaching suggestions: event data cannot establish off-ball spacing or prove the cause of a defensive problem.
What's next
Test the full workflow with nontechnical coaches, support their own match records, improve contextual review, and let coaches revisit a chosen focus after their next game. The current demo uses historical tournament data; training ideas are proposals for coach judgment, and improved team performance has not yet been validated.
Log in or sign up for Devpost to join the conversation.