Wang Qiur: Serve Better

Overview

Elite serve analysis has historically been locked behind $10,000 systems like PlaySight or $50K/year IMG Academy programmes. Wang Qiur: Serve Better is a smartphone-based AI coach that delivers real-time biomechanical feedback on the tennis serve — the most complex, high-ROI shot in the game — to any player, anywhere, at any time.

The project was built as part of an HKU multi-disciplinary team, with Jonathon Simpson & Co. contributing the core technology and AI development.

The Problem

Tennis players attend a lesson on Tuesday and forget everything by Saturday. There is no affordable, objective feedback loop for solo practice. Human coaches give subjective cues (“hit harder,” “more spin”) while players want “Your arm is 12° off pro average.” For grassroots and intermediate players, that level of analysis simply didn’t exist.

The Solution

A web-based serve analysis tool that runs entirely in the browser with no app download required. Players either upload a video or use live webcam mode. The system:

• Tracks 33 skeletal landmarks using MediaPipe pose estimation
• Detects ball and racket using a custom-trained YOLOv8+ model at 60fps
• Evaluates 18 biomechanical metrics across serve phases (load, trophy, contact, follow-through)
• Benchmarks against pro serve data and delivers severity-graded, phase-by-phase coaching findings
• Generates AI drill recommendations via OpenAI API
• Provides real-time text-to-speech audio feedback — no need to look away from the court

Key Technical Achievements

• Skeleton tracking fully functional with an average 83% detection ratio across 21 saved runs; 16 of 21 runs exceeded 95%
• Custom-labeled dataset of 4,013+ image-label pairs (manual + auto-labeled) for ball and racket tracking
• Stable export pipeline: keypoints JSON, summary JSON, metrics CSV, and overlay video
• Shadow Swing Mode — analyse serve mechanics without a ball, enabling indoor practice anywhere
• Pro-mirror video overlay: compare your serve skeleton directly against a professional’s

Competitive Edge

Unlike SwingVision (iOS-only, result-oriented) or MWM Pro (general findings), Serve Better is the only tool offering serve-specific biomechanical depth with phase detection, severity-graded coaching, and browser-based access on any device.

Role

Technology and AI development — skeleton tracking pipeline, YOLO dataset preparation and training, web integration, OpenAI API interpretation layer, and export pipeline.

---

Outcome: Delivered a functional AI-powered biomechanical serve analysis tool with 83% average skeleton detection ratio, bringing elite-level coaching to any smartphone at 1/100th the cost of professional systems.