My AI Agent Wrote All the Code for an Orienteering Analyzer. The Hard Part Came After It Worked.

My daughter participates in orienteering competitions almost every week. After each competition, I analyze her performance, focusing on where time was lost, why it happened, and if the same issues occur consistently. Traditionally, I used another service for this analysis, but it fell short in georeferencing the map, marking control points, cutting GPS tracks into splits, and comparing each leg with other participants' results.

To address this, I developed a portal for orienteering analysis. This portal has an unusual feature: an AI agent wrote all of its code. I never saw any of the code, but I provided tasks, reviewed the results in a browser, tested them against real competitions, and generated subsequent tasks.

The portal requires three main inputs in various formats: a scanned map image, a GPX track from a watch, and an official race protocol. The goal is to transform these inputs into valuable analysis. First, the map image is georeferenced using at least three matching points on the orienteering map and a base map. This transformation allows the digital route and control points to be created.

Next, the GPX track is divided into splits based on the control point coordinates. Each split is analyzed for start, end, time, track length, and pace. Although GPS is not perfect, the portal allows for manual corrections in areas where data may be messy.

Once the data is prepared, the portal compares each leg with the official protocol, identifying where the athlete performed well or struggled. This comparison creates an "Ideal Leader" benchmark, showcasing the best times for each split from various participants rather than just the winner. The portal highlights problematic sections and allows for detailed analysis of individual splits.

Finally, when focusing on a specific problematic leg, the portal provides a visual representation of the map, track, control points, split parameters, and an AI-assisted analysis. The AI receives a context-specific prompt including a PNG of the relevant map area, split numeric parameters, and generates targeted insights to guide the athlete's improvement.