Around 300 million kids across the world play football in some form today – whether in training or just for fun. But the path from playing on a local pitch to a professional career is incredibly narrow.
The problem lies in the numbers: on average, there’s only one scout for every 2,000 potential talents. In elite regions, that ratio might drop to 1:500, but in rural areas of developing countries, it can be drastically higher. The question becomes: how are we going to discover the next Messi and Ronaldo?
The 1:2000 Challenge
Traditional scouting systems have obvious limitations. A human scout can’t be everywhere – logistics, geography, and time create insurmountable barriers. An AI-driven system offers a solution: it’s always accessible, faster, and consistent in its measurements.
Our team at Levi9 worked on developing an application that puts opportunity directly into young athletes’ pockets. The idea was simple but powerful: young talents perform set drills, record them with a dedicated app on their smartphones, and AI analyzes the footage and returns scores and suggestions. It no longer matters whether a scout can physically attend a game – the system is always there.
Scaling Under Pressure
The most important technical challenge was architecting a solution able to efficiently scale on demand while maintaining excellent performance with minimal cost increase. The solution also needed to allow rapid development of new features, support further changes to the system, and not be tied to a single Cloud provider.
The team used many proven tools and services to build the system. However, they also developed many bespoke solutions that made it shine even at peak loads that were 40 times higher than regular use. A prime example is the AIOps-powered Observability system that utilizes real-time monitoring, anomaly detection, alert correlation and triage, and root cause analysis to keep the system stable.
Teaching AI to See Like a Scout
At the heart of the application lies sophisticated AI/ML technology. Computer Vision is used to automatically detect and track people and objects like balls, bats, clubs, and racquets. The proprietary 3D Pose Engine uses deep learning models to extract full-body skeletal and object key points – including up to 20 joint centers per frame.
The system combines metrics about body position, velocity, accelerations, and center of mass to understand motion and analyze performance as it happens. This enables precise and objective evaluation of technique and execution in real-time.
Testing by Doing: Why We All Grabbed a Ball
The project required combining cutting-edge technology with deep experience from previous ventures. The team – a diverse mix of mobile, web, backend, cloud, AI, and QA experts – was eager to step into all kinds of unknowns and support each other while keeping the mission on track.
Testing was particularly hands-on: even the most dedicated “desk sitters” on the team were soon grabbing footballs and attempting drills, struggling to get high scores without hurting themselves in the process. This hands-on approach ensured that the system worked not just in theory, but in real-world conditions.
Calm Waters, Bold Experiments
The team was fortunate to build the majority of this amazingly interesting concept from scratch. This allowed them to combine all previous experiences and expertise to produce strong foundations for further expansions while achieving stability and gaining trust.
After a smooth initial release, the team found themselves in calm waters and confident to experiment. They explored white-labeling of the product, cloud portability, AI search with text and voice, custom camera and recording implementation with on-device AI capabilities, boosting user engagement with interesting integrations and personalization, and many more innovations.
AI as a Tool, AI as a Mindset
From implementations of the core AI/ML engine, through AIOps-powered observability and scaling, to AI-designed and spec-driven AI-coded apps and features, AI is an integral part of both the product being built and the software development lifecycle itself.
This project demonstrates how AI expertise goes beyond just implementing algorithms – it’s about creating systems that scale, adapt, and continuously improve while solving real-world problems that impact millions of young athletes worldwide.
***This article is part of the AI9 series, where we walk the talk on AI innovation.***
