VisionSelect AI

VisionSelect AI uses Gemini-powered video analysis to evaluate cricket players with objective performance insights, reducing selection bias and helping coaches discover talent through explainable AI.

Comment

About the Project

Inspiration

VisionSelect AI was inspired by a simple but important problem in cricket: talent selection can sometimes depend too much on human judgment, limited observation, and unconscious bias. A talented player may be overlooked because a scout cannot watch every match or because traditional evaluation focuses heavily on a few visible statistics.

We wanted to build an AI-powered platform that could help coaches and selectors make more objective, explainable, and evidence-based decisions. Instead of simply giving a player a score, VisionSelect AI is designed to explain why the player received that score and which aspects of their performance influenced the decision.

What We Built

VisionSelect AI is an AI-powered cricket talent evaluation platform that analyzes match footage and converts player performance into meaningful insights.

The platform allows users to:

  • Upload cricket match videos for analysis.
  • Detect and evaluate player performance.
  • Generate Batting, Bowling, and Fielding performance scores.
  • Produce an overall AI Player Score.
  • Explain “Why this score?” using performance-based evidence.
  • Compare multiple players.
  • Generate coach-friendly reports.
  • Provide a Fairness & Bias Transparency view.
  • Allow human corrections so that coaches can remain part of the decision-making process.

Our goal was not to replace scouts or coaches, but to create an AI assistant for selectors that makes their evaluation process faster, more consistent, and more transparent.

How We Built It

We developed VisionSelect AI using a modular architecture:

  • Frontend: React, TypeScript, Vite, Tailwind CSS, and Framer Motion.
  • Backend: Spring Boot with Java for authentication, APIs, video management, and application logic.
  • AI Service: Python with computer-vision and machine-learning components such as OpenCV, MediaPipe, and YOLO.
  • Database: PostgreSQL for storing users, players, videos, AI jobs, analysis results, and reports.
  • Generative AI: Google Gemini API for generating meaningful and explainable insights from the analyzed performance data.

The workflow is designed as a pipeline:

Video Upload → AI Processing → Player Performance Analysis → Explainability → Results Dashboard → Player Comparison → Coach Report

We also designed the system with security and scalability in mind, including JWT-based authentication, role-based access control, asynchronous video processing, and separation between the frontend, backend, and AI services.

What We Learned

Building VisionSelect AI taught us that an AI product is much more than simply connecting an AI model to an application.

We learned how to:

  • Design a complete AI-powered full-stack architecture.
  • Integrate React, Spring Boot, Python, PostgreSQL, and Gemini into one system.
  • Work with large video-processing workflows.
  • Build REST APIs and secure them using JWT authentication.
  • Design asynchronous AI-processing pipelines.
  • Use computer vision for extracting meaningful information from video.
  • Store and present AI-generated results in a user-friendly way.
  • Think about AI explainability and fairness, rather than focusing only on prediction accuracy.
  • Keep a human-in-the-loop so that AI recommendations can be reviewed and corrected by coaches.

One of our biggest lessons was that trust is as important as accuracy. If an AI system recommends a player, users should be able to understand the evidence behind that recommendation.

Challenges We Faced

One of the biggest challenges was designing a system capable of handling video analysis, because video files are much larger and more computationally expensive than normal application data.

We also faced challenges in connecting multiple technologies into a single reliable workflow. The frontend, Spring Boot backend, Python AI service, database, and Gemini integration each have different responsibilities and communication requirements.

Another major challenge was AI explainability and fairness. A simple score is not enough for a real-world selection system. We therefore focused on showing the factors contributing to a player's evaluation and designing a fairness layer to identify potential bias in selection outcomes.

We also had to think carefully about authentication, authorization, video-upload security, asynchronous processing, error handling, and how to keep the application responsive while AI analysis is running.

Why VisionSelect AI Matters

VisionSelect AI is built around the idea that technology should assist human decision-making, not hide it behind a black box.

By combining computer vision, machine learning, generative AI, and a human-in-the-loop approach, VisionSelect AI aims to give coaches and selectors a more transparent way to discover and evaluate cricket talent.

Our vision is to make talent identification faster, fairer, explainable, and accessible, so that promising players have a better chance of being recognized based on their actual performance.

Built With

Share this project:

Updates