AI Transparency

Understanding how our AI analyzes your swimming technique

How It Works

Upload Video → Pose Detection → Metric Calculation → AI Analysis

Step-by-Step Process

  1. Video Upload: Your video is uploaded securely to our server for processing
  2. Frame Extraction: Individual frames are extracted from the video
  3. Pose Detection: MediaPipe AI identifies 33 body landmarks in each frame
  4. Metric Calculation: Algorithms analyze landmark positions to calculate swimming metrics
  5. Stroke Classification: Pattern recognition determines your swimming stroke type
  6. AI Coaching: Your metrics are sent to an AI model to generate personalized advice

Technologies Used

Pose Detection

Google MediaPipe Pose

Industry-leading pose estimation that detects 33 body landmarks in real-time. Used by researchers and developers worldwide.

AI Coaching

Google Gemini / LM Studio

Large Language Model that interprets your metrics and provides personalized coaching advice based on swimming best practices.

Video Processing

OpenCV

Open-source computer vision library for video frame extraction, processing, and annotation.

Analysis Engine

Custom Algorithms

Proprietary algorithms calculate stroke rate, body angles, and efficiency metrics from detected poses.

Data Collection & Privacy

What We Collect

Data Type Purpose Storage
Video File Analysis processing Temporary - deletable anytime
Swimming Metrics Results display & coaching Stored with job ID only
Chat History Conversation continuity Linked to job ID, not user
✓ No Account Required: Use the service without creating an account or providing personal information
✓ No Tracking: We don't use tracking cookies or collect browsing data
✓ Video Privacy: Only metrics (not video content) are sent to AI for coaching
✓ Deletable: Delete your videos and analysis data anytime via History tab
✓ No Sharing: Your data is never sold or shared with third parties
✓ Local Option: Use LM Studio for fully local AI coaching (no data leaves your network)

How Metrics Are Calculated

Stroke Rate

Method: Tracks shoulder joint movement to detect complete stroke cycles. Counts cycles and divides by time to get strokes per minute.

Accuracy Factors: Clear visibility of arms, consistent camera angle, minimal splash interference.

Body Alignment

Method: Calculates the angle between shoulder and hip landmarks relative to horizontal. Averaged across all valid frames.

Accuracy Factors: Full body visibility, side-view camera angle, consistent lighting.

Distance Per Stroke (DPS)

Method: Estimates forward displacement between frames using body landmarks. Uses average torso length (0.5m) as scale reference.

Note: This is an estimate - actual pool distance measurements will be more accurate.

SWOLF Score

Method: Calculated as Time (seconds) + Stroke Count. Based on video duration and detected strokes.

Note: For accurate SWOLF, use a known distance (e.g., one pool length) video.

Stroke Classification

Method: AI analyzes arm movement patterns, body rotation, and kick style to classify stroke type (freestyle, backstroke, breaststroke, butterfly).

Correction: If misclassified, tell the AI coach "I was swimming [correct stroke]" to get adjusted advice.

Accuracy & Limitations

Important Disclaimers

  • All metrics are estimates based on video analysis, not direct measurements
  • Results may vary significantly based on video quality, camera angle, and conditions
  • This tool is for training insights only - not official competition timing
  • AI coaching is general advice - not a replacement for a qualified coach
  • This is not medical advice - consult professionals for health-related questions

Factors Affecting Accuracy

Factor Impact Recommendation
Camera Angle High Side view at pool deck level
Video Resolution Medium 720p or higher
Lighting Medium Well-lit, avoid backlighting
Water Clarity Low-Medium Clear water preferred
Splash/Bubbles Low Some interference is acceptable

AI Coaching Explained

How AI Coaching Works

  1. Context Preparation: Your swimming metrics are formatted into a structured summary
  2. Expert Knowledge: The AI is given context about swimming best practices and technique
  3. Analysis: The AI interprets your specific metrics to identify strengths and areas for improvement
  4. Personalized Advice: Recommendations are generated based on your unique data
  5. Conversation Memory: The AI remembers previous messages for coherent follow-up

What the AI Can See

  • Your swimming metrics (stroke rate, SWOLF, body angle, etc.)
  • Your previous chat messages in the current session
  • General swimming technique knowledge

What the AI Cannot See: Your actual video, personal information, or any data from other users.

AI Coaching Limitations

  • Cannot see your actual video - works only from numerical metrics
  • May occasionally misinterpret unusual metric combinations
  • Provides general best practices, not individualized medical advice
  • Cannot guarantee improvement - results depend on training application
  • Should supplement, not replace, guidance from qualified coaches

Your Rights & Control

🗑️ Right to Delete: Delete your videos and analysis data anytime via the History tab
📥 Right to Export: Export your metrics data via our API
🔒 Right to Privacy: Use locally with LM Studio for zero external data sharing
❓ Right to Understand: This page explains exactly how your data is used

Questions?

If you have questions about our AI, data practices, or how the analysis works, please: