Machaxi and SoftwareOne explore AI-powered sports coaching with Claude

Exploring the next generation of AI-assisted coaching

The successful badminton app uses Claude – but SoftwareOne was not involved in that aspect of the build. Instead, Machaxi sought SoftwareOne’s expertise to help it capture fast, subtle wrist movements without specialized cameras and overlay an ideal, AI-generated skeletal pose on the badminton player’s movement. “We saw value and merit in what they [SoftwareOne] had to say to us,” says Pratish Raj, co-founder and CEO of Machaxi.

The badminton application has since been launched for Machaxi’s beginner coaching users. A player uploads a video, the custom model interprets body posture, angles and game context, and Claude helps express the analysis as understandable voice feedback.

The successful advisory engagement with SoftwareOne gave Machaxi the confidence to involve SoftwareOne much more extensively in its next ground-breaking product.

Choosing Claude for reasoning, planning, and feedback

The second project extends Claude’s role from providing consistent post-session feedback to delivering guidance in real time.

The product is being designed to work with each user’s fitness history and Claude is being evaluated across reasoning, planning and personalized feedback workflows. Together, they create an appropriate program and guide the user through it by voice. “In fitness, sports, health, et cetera, we have not yet seen a product which is fully agentic AI in its capability, and that’s what we are attempting to build here,” Raj says.

The teams are also researching how AI can generate timely, useful coaching feedback. “For this ambitious project using Claude, SoftwareOne is partnering with us in the true sense of partnership,” Raj says. “There is strong alignment on priorities, with both teams working closely together to drive progress and support a successful go-live.”

The prototype architecture combined Amazon Nova Sonic for conversational voice interactions, computer vision models for movement analysis, Amazon Kinesis and S3 for data streaming and storage, and Claude as the reasoning layer. Claude evaluated movement data alongside personal context to generate tailored coaching recommendations and feedback.

The joint team did not choose Claude by default. It evaluated a broad range of models available through Amazon Bedrock and ran additional tests with Google Gemini and OpenAI’s GPT models. Raj particularly values Claude’s ability to examine the premise behind a request. “Just like a human would, it questions me back on my assumptions,” he says. “And that’s very important in a conversation.”

Detailed post-session analysis is structured so that the voice experience can use in future sessions, creating a longitudinal memory of the user’s development. The coach can therefore remind the user of a previous problem and help prevent it from recurring.