This is a submission for the Google AI Studio Multimodal Challenge
🥊 What I Built
I built the MMA Coach Assistant — an AI-powered web application that transforms raw fight footage into actionable coaching intelligence. This tool solves a critical problem in combat sports: the lack of affordable, instant, and objective fight analysis for fighters, coaches, and academies.
Instead of spending hours manually reviewing tapes or hiring expensive analysts, users simply upload a video, and within seconds, receive:
- Quantitative performance metrics (strike accuracy, takedown success, control time)
- Qualitative tactical insights (“drops left hand after right cross”)
- Head-to-head fighter comparisons
- Personalized 7-day training plans
- Integrated e-commerce for official fighter merch
This isn’t just a video analyzer — it’s a complete digital coaching platform that democratizes access to elite-level fight breakdowns using AI.
🎥 Demo
🔗 Live App (Deployed on Cloud Run):
👉 https://mma-coach-assistant-78710804281.us-west1.run.app/#/coach
🎬 Video Walkthrough (YouTube):
▶️
📁 GitHub Repository (Open Source MIT):
🐙 https://github.com/PauloTuppy/MMA-Coach-Assistant
(Note: Since I used Google Gemini 2.5 Flash for video understanding — which remains available — the live app is fully functional during judging.)
🧠 How I Used Google AI Studio
I leveraged Google AI Studio as the core AI engine for all video analysis and insight generation. The entire multimodal pipeline was prototyped, tested, and optimized within Google AI Studio’s intuitive interface before being integrated into the production app via API.
Key steps:
- Prompt Engineering: Designed structured prompts to extract specific metrics and insights from fight videos (e.g., “Identify all jabs, crosses, takedowns, and positional control moments. Calculate strike accuracy and control time.”).
- Model Selection: Chose Gemini 2.5 Flash for its speed, cost-efficiency, and strong multimodal video understanding — perfect for real-time fight analysis.
- API Integration: Connected the Google AI Studio API to my React frontend and Node.js backend to send video files (or URLs) and receive structured JSON responses containing metrics, insights, and summaries.
- Testing & Iteration: Used Google AI Studio’s playground to rapidly test prompts against real fight clips, refining output quality and consistency.
🎯 Multimodal Features
I implemented Gemini 2.5 Flash’s video understanding capabilities to create a truly multimodal experience:
1. Video → Metrics & Insights
- The AI watches the uploaded fight video and extracts quantitative data (e.g., 88% strike accuracy, 3 successful takedowns).
- It then generates qualitative coaching feedback (e.g., “Struggles to get up from bottom position,” “Excellent jab for distance control”).
✅ Why it enhances UX: Turns passive video watching into active, data-driven coaching — saving hours and removing human bias.
2. Video → Fighter Comparison
- By analyzing both fighters’ movements, the AI produces a side-by-side tactical breakdown, highlighting strengths, weaknesses, and keys to victory.
✅ Why it enhances UX: Enables coaches to prepare game plans by understanding opponent tendencies — all from a single video.
3. Video → Personalized Training Plan
- Based on the identified weaknesses (e.g., poor get-up defense), the AI generates a customized 7-day training schedule with specific drills and focus areas.
✅ Why it enhances UX: Transforms analysis into immediate, actionable improvement — closing the loop between data and development.
4. Video → Narrative Fight Summary
- The AI synthesizes the entire bout into a coherent, contextual summary (e.g., “Classic striker vs. grappler matchup…”).
✅ Why it enhances UX: Helps users quickly grasp the strategic narrative of the fight, not just isolated stats.
💡 Innovation & Impact
This applet goes beyond a simple demo — it solves a real-world problem with measurable impact:
- For Fighters: Faster feedback, targeted training, accelerated growth.
- For Coaches: Objective data, strategic clarity, time savings.
- For Gyms: Affordable analysis tool, improved coaching quality, merchandise revenue.
By combining multimodal video understanding with personalized output generation, the MMA Coach Assistant showcases the true power of Gemini: not just to see, but to understand, interpret, and advise.
👥 Team Submission
Solo submission by Paulo Tuppy.
Built with ❤️ for fighters, by a developer who believes AI should empower, not replace, human coaching.
Top comments (0)