Job description
We are developing a browser-based virtual jewelry try-on application that allows users to see realistic 3D rings on their fingers using a phone or computer camera.
We already have a working 3D implementation built with MediaPipe, Three.js and React. The application can track the hand, position a 3D ring on the finger, handle basic occlusion and render jewelry in real time.
However, we want to significantly improve the accuracy and realism of the current implementation, particularly:
Ring centering and positioning on the finger
Accurate orientation as the hand/finger rotates
Correct fit and depth
Stability and reduction of jitter
Side-angle accuracy
Occlusion
Performance across different hand positions
Consistency between front and rear mobile cameras
We are looking for a highly experienced computer vision / AR engineer to review our existing implementation, identify the primary limitations, and determine what algorithmic improvements could bring the tracking quality substantially closer to leading commercial virtual jewelry try-on solutions.
This is not a project starting from scratch. You will be reviewing and improving an existing application and working alongside our current developer.
For the initial phase, we would like a technical assessment of the current approach followed by implementation/testing of the highest-impact improvements. If the initial work produces promising results, this can develop into a larger ongoing project.
Strong experience in some of the following is important
Computer vision
MediaPipe / hand landmark tracking
Hand pose estimation
3D geometry and coordinate systems
AR object placement
Pose estimation and transformation
Temporal filtering / tracking stabilization
Depth estimation
Three.js / WebGL
Real-time mobile/browser computer vision
Occlusion / segmentation
Experience with virtual try-on, AR commerce, hand tracking, rings/watches/jewelry, or similar camera-based object placement would be especially valuable.
When applying, please answer:
What experience do you have with hand tracking, pose estimation, AR or virtual try-on?
Have you worked with MediaPipe hand landmarks? If so, please describe the project.
Have you previously improved stability/accuracy beyond the raw output of an existing tracking framework such as MediaPipe?
Based on the description above, what areas would you investigate first to improve ring positioning, orientation and stability?
Please provide examples of relevant computer vision/AR work.
We are specifically looking for someone with deep technical experience, rather than a general web or 3D developer.
• ****Because this project involves access to proprietary source code, selected developer must be currently residing and performing work in the United States. Please confirm this in your proposal.