Computer Vision Data Annotation Platform

Product challenge

Computer-vision customers needed accurate training data from images and point clouds, while distributed annotators needed tools that remained understandable and responsive during detailed, repetitive work. Better product ergonomics directly affected quality, throughput, and delivery cost.

Product and engineering approach

• Built the interaction and state foundations with React, Redux, and a unified model for complex annotation sessions.
• Used Konva for precise 2D canvas work and Three.js for point-cloud rendering, creating custom controls where standard interactions were not accurate enough.
• Optimized rendering and data processing for large datasets so technical limitations did not interrupt the annotator’s task.
• Designed workflows for distributed contributors, connecting interface clarity and engagement to the platform’s throughput requirements.
• Worked close to the domain and users, which later enabled me to move into product management for the same toolset with practical implementation context.

Outcome

• Delivered one platform for complex 2D image and 3D point-cloud annotation workflows.
• Improved precision through purpose-built controls and improved responsiveness on demanding datasets.
• Made difficult annotation work more usable for a remote workforce, supporting both data quality and productivity.