Computer vision engineer needed *mediapipe*

Summary **Computer Vision Exercise Engineer — Pose Estimation / MediaPipe / Blazepose** **This is a long-term, per-deliverable contract. Read carefully before applying.** We are a funded fitness technology company building an AI-powered health and workout tracking app. We are expanding our exercise library using the QuickPose iOS SDK (built on MediaPipe BlazePose) and are looking for a technically skilled developer to define, implement, test and document exercises at scale. This is not a general development role. This is a highly specific, repeatable technical task. If you do not have hands-on experience with MediaPipe, BlazePose or pose estimation SDKs, do not apply. --- **What You Will Be Doing** For each exercise you will: 1. Implement the exercise rep counter using the mediapipe 2. Define all joint angle thresholds, rep boundary conditions and state machine logic 3. Define form feedback rules (what triggers a form warning and under what landmark conditions) 4. Store all definitions in our provided database schema (JSON format provided) 5. Record video evidence showing correct rep counting with the mediapipe overlay active 6. Document edge cases — partial reps, camera angle variation, occlusion 7. Submit a landmark data sample (minimum 3 complete reps) in our schema No submission is accepted without video evidence. No video, no payment. This is non-negotiable. --- **What You Must Have** - Demonstrated experience with MediaPipe, BlazePose. - Understanding of skeletal keypoint data (33-point body model) - Understanding of joint angle calculation from landmark coordinates - Ability to define rep counting logic programmatically - Basic sports science or kinesiology knowledge — you must understand what correct exercise form looks like - Swift or ability to work within a fast moving environment - Strong written English for documentation **What You Must NOT Have** - No experience with pose estimation = do not apply - No portfolio showing relevant work = do not apply - If your plan is to figure it out as you go = do not apply --- **Deliverable Format** Every exercise submission must include: - Completed definition document (template provided) - Video recording with QuickPose overlay showing accurate rep counting - Edge case notes - Raw landmark JSON sample (3 reps minimum) - MET value with source citation Incomplete submissions will not be reviewed. --- **Pay Structure** This is a per-exercise rate, not hourly. - Simple exercises (single joint, 2-phase movement): $X per exercise - Medium exercises (multi-joint, 3-phase movement): $X per exercise - Complex exercises (full body, multi-directional): $X per exercise We have approximately 80 exercises scoped. This is a long-term engagement for the right candidate. Fast, accurate, high quality work will be rewarded with consistent volume. Payment is released upon verification of video evidence and schema compliance. Typically within 24 hours of submission. --- **How to Apply** Do not send a generic proposal. Applications without the following will be ignored: 1. One example of previous pose estimation or MediaPipe work (GitHub, video or portfolio link) 2. A list of 5 exercises you are confident you could implement immediately 3. Your estimated timeline to complete 10 exercises per week 4. One sentence explaining what a MediaPipe landmark index is We will begin with a paid trial of 3 exercises at full rate. If quality meets our standard we will proceed with the full scope immediately. We are looking for one person only. We are looking for the right person, not the fastest applicant with a long term; high paying role at the end.