VoiceGym is an AI-driven fitness coach inspired by the need for accessible, professional-grade training, using MediaPipe and Google Gemini to provide real-time form analysis and multilingual voice feedback for 8+ exercises. We built the platform using a Python backend and PostgreSQL for scalable progress tracking, overcoming significant challenges in real-time latency and concurrent audio-visual processing. We are proud of achieving high-accuracy biomechanical analysis and a seamless user experience, which taught us the complexities of high-frequency data pipelines. Moving forward, we plan to expand VoiceGym with IoT wearable integration and advanced AR workout overlays to further revolutionize the home fitness experience.