Monday, February 5, 2024

AI-Curious, Episode 2 - Building an AI Business, with SwingVision CEO Swupnil Sahai

  

Serving Up Smarter Tennis with AI

Former Tesla Engineer's App Uses AI to Improve Your Game

Swupnil Sahai first got the idea for Swingvision, an AI-powered tennis tracking app, because he wanted better stats on his own game. As an engineer who had worked directly under Elon Musk on Tesla's self-driving car AI, Sahai wondered if AI could be applied to tennis as well. The resulting app uses computer vision and neural networks to record matches, provide advanced stats, create highlights, and even make line calls more accurately than the human eye.


1. 11:45, Early Inspiration at Tesla

Sahai shares how working on Tesla's single-camera self-driving car approach showed him the possibilities of computer vision AI to track objects. Discussions with Musk also inspired Sahai to find simplified, affordable solutions rather than rely on expensive sensor hardware.


2. 15:30, How Neural Networks Work

Sahai explains how neural networks automatically find visual features correlated with particular tasks, removing the need for human domain experts to manually identify important elements. This allows the AI to determine on its own what factors matter most for tracking tennis balls.


3. 19:40, Building a Data Model

In order to train the AI, Sahai and his team recorded amateur tennis matches, scraped YouTube for old Federer and Djokovic footage, and captured video from top college teams. This real-world visual data was essential for the neural network to learn to track tennis balls.


4. 23:30, AI Strengths and Weaknesses

Surprisingly, the AI excels at close line calls within 10 cm but struggles to track obvious outs like balls hit far wide or over the fence. Sahai explains this is due to lack of model training on rare edge case data rather than limitations in the core approach.


5. 32:11, Future AI Applications

Sahai sees major democratization of film production as AI generators make high-quality special effects and video content available to all at low cost and without a Hollywood budget. He also expects AI assistants to surface relevant personal information from recordings of our digital lives.


6. 36:00, AI Startup Advice

For those working in AI, Sahai advises focusing first on solving real customer problems rather than attachment to particular technical solutions. He also recommends building novel datasets for defensible, vertical AI businesses rather than relying solely on APIs.


So while we probably shouldn't expect AI to hit a two-handed backhand anytime soon, Swingvision shows there's serious potential for smarter sports tech and many more hidden applications waiting to be uncovered through entrepreneurial creativity. Game, set, and match to AI!


No comments:

Post a Comment

2024 MLP Playoffs Dallas - Sorry Not Sorry Recap

  0:00 Michael Loyd Getting Hit In The Head 2:09 Frisco Pandas vs. Miami Pickleball Club (1) 8:31 Frisco Pandas vs. Miami Pickleball Club...