HIITSurvey

Inspiration

Surveys are a nuisance. That's what people say when they don't use our platform. We breed competition and produce winners through survey and fitness. You're a software engineer rotting at your desk all day. You need a damn break. Why not spice up your day by getting a quick dopamine rush from some exercises while supporting a good cause?? Even better, you can compete against your friends to see who gets the most points. Eventually either you give in or your ego does.

What it does

HIITSurvey is a gamified fitness web application that transforms survey completion into an engaging physical activity. Users complete surveys from SurveyMonkey while performing exercises (push-ups, squats, jumping jacks, arm circles) tracked in real-time through their webcam.

Key Features:

Real-time Exercise Detection: Uses MediaPipe Pose Landmarker to count exercises from webcam video (~10 FPS) with custom computer vision algorithms
Auto-advancing Surveys: Questions automatically advance after completing 5 reps of the mapped exercise, creating a seamless flow between physical activity and survey responses
Retro Arcade UI: Pixel-art inspired interface with nostalgic gaming aesthetics
Comprehensive Rewards System: Points, streaks, badges, and leaderboards to drive engagement
Social Competition: Leaderboard integration via Supabase to compete with friends
AI-Powered Workouts: OpenAI integration generates personalized workout plans
Text-to-Speech: ElevenLabs integration provides audio guidance during exercises

How we built it

HIITSurvey is a full-stack web application built with modern technologies:

Frontend:

React 19 with Vite for fast development and optimized builds
Real-time webcam capture (~10 FPS) using MediaStream API
Retro arcade-style UI with pixel-art aesthetics
React Router for navigation
Supabase client for authentication and leaderboard data

Backend:

FastAPI with Python 3.14.2 for high-performance async API endpoints
MediaPipe Pose Landmarker extracts 33 pose landmarks from video frames
Custom computer vision algorithms analyze joint angles and distances to detect:
- Push-ups (elbow angle tracking)
- Squats (knee and hip angle analysis)
- Jumping jacks (arm and leg distance measurements)
- Arm circles (wrist trajectory tracking)
- And 7 additional exercise types
Modular architecture with separated routers, services, and models

Integrations:

SurveyMonkey API: Full workflow integration for fetching surveys and submitting responses
OpenAI API: Generates personalized workout plans based on user preferences
ElevenLabs API: Text-to-speech for audio exercise guidance
Supabase: User authentication, leaderboards, and data persistence

Architecture Highlights:

Frame-by-frame processing: Frontend sends video frames to backend for pose detection
State management: Exercise counters tracked server-side with reset capabilities
Real-time feedback: Immediate rep counting and exercise detection feedback
Error handling: Robust error handling for API failures and edge cases

Challenges we ran into

Computer Vision Algorithm Development: Despite MediaPipe providing accurate pose landmarks, writing reliable algorithms to identify exercises proved extremely challenging. We had to:

Account for varying camera angles and user positions
Handle partial occlusions and pose estimation errors
Implement state machines to track exercise phases (up/down, open/closed)
Tune thresholds for different body types and movement speeds
Prevent false positives from similar movements

Real-time Performance: Processing video frames at 10 FPS while maintaining low latency required careful optimization:

Balancing frame rate with processing time
Managing state across multiple concurrent exercise detections
Handling network latency between frontend and backend

SurveyMonkey Integration: Implementing a complete end-to-end workflow with SurveyMonkey's API required:

Understanding complex survey structures and question types
Mapping exercises to survey questions dynamically
Handling authentication and rate limiting
Ensuring data integrity when submitting responses

State Synchronization: Keeping exercise counters, survey progress, and rewards synchronized across the frontend and backend while maintaining a smooth user experience.

Accomplishments that we're proud of

Accurate Exercise Detection: Built reliable computer vision algorithms that accurately detect 11 different exercise types using only pose landmarks—no machine learning models required, just pure geometric analysis.

End-to-End SurveyMonkey Integration: Successfully implemented a complete workflow from survey fetching to response submission, with automatic question advancement based on exercise completion.

Polished User Experience: Created an engaging retro arcade-style UI that makes survey completion feel like a game, complete with rewards, badges, streaks, and leaderboards.

Real-time Performance: Achieved smooth real-time exercise tracking at 10 FPS with minimal latency, providing immediate feedback to users.

Comprehensive Rewards System: Built a sophisticated gamification system with points, streaks, duration bonuses, milestone badges, and social competition through leaderboards.

What we learned

Traditional Computer Vision Still Has Value: While machine learning models dominate modern CV, we discovered that well-designed geometric algorithms using pose landmarks can be highly effective for exercise detection. Understanding joint angles, distances, and movement patterns allowed us to build accurate detection without training models.

The Power of Gamification: Adding game mechanics (points, badges, streaks, leaderboards) transforms mundane tasks into engaging experiences. The psychology of competition and achievement significantly increases user motivation.