System Architecture

Home Run to Mars (BaseballTrainer)

Home Run to Mars is an AI-powered baseball swing analysis platform that delivers real coaching feedback from just a video clip — designed to feel fun, motivating, and accessible for every player.

🎯 What It Does

Users upload or record their swing, and within seconds, the system:

Extracts key body, bat, and ball motion using MediaPipe Pose + YOLOv8.
Identifies swing phases, detects contact, and calculates metrics like bat speed, exit velocity, and launch angle.
Sends high-value frames to Anthropic Claude (via OpenRouter) for instant, personalized coaching feedback.
Uses Google Gemini to generate custom drill recommendations based on the player's needs.
Converts feedback into a motivating coach-style voice-over through ElevenLabs.
Rewards improvement with a space-themed progression system (Atmosphere → Orbit → Moon → Mars).

🏗️ Architecture

We use a microservices architecture optimized for clarity, scalability, and maintainability:

Frontend: Next.js + TypeScript + Tailwind + Framer Motion
Gateway: Node.js/Express for routing, Auth0 JWT verification, uploads
Pose Analysis Service: Flask + MediaPipe + YOLOv8 + OpenCV + Kalman Filters
Drill + Sensor Services: Independent Flask microservices
AI APIs: OpenRouter (Claude), Google Gemini, ElevenLabs TTS
Database: MongoDB Atlas (cloud)
Local Uploads → Cloud Storage Ready (S3 planned)

🚀 Why It’s Special

Where most tools stop at pose keypoints, we combine full biomechanics, instant AI coaching, and gamified progression. It’s fast enough to engage beginners and accurate enough to impress coaches — all while making improvement fun.

Training doesn’t have to feel like work. We make it feel like a mission to Mars.