Context
Ingedata was founded in the early 2000s as an IT services company. By the late 2010s, the company decided to pivot to data labelling, doing all sorts of data annotation but focusing mostly on image annotation.
The problem
Offshore services abound in Madagascar, but data annotation for AI was clearly a blue ocean. The strategy was perfect. But we had to build the platform to work on, and the back-office to track the work and upload the data batches to annotate.
My work
I created the back-office for KPI tracking and batch processing from scratch. I chose Ruby on Rails on the back end and React, reinforced with TypeScript, on the front end. The auth feature and most of the UIs were standard, but batch processing was where I had to be diligent about queue management. The upload had to stay smooth without stopping users from doing other work.
I worked on the early-stage, canvas-based annotation platform with an international team of five English-speaking engineers. The project was built largely in TypeScript, with controls built in Vue.js. It was precision work: a deviation of even a few pixels could distort the data. We had to load heavy JSON data onto images and keep that accuracy through zoom, unzoom and image navigation.
I also worked on an early implementation of OpenCV grabcut on the tool in order to increase selection speed on some use cases.
The results
- Around 200 operators use the labelling platform
- Diverse annotation projects, including healthcare, geospatial and waste management, have been completed for client companies across Europe and Asia
- The dashboard became the daily instrument for tracking annotation operations
All images are the property of Ingedata and are shown for illustration only. The pictures above can be viewed on the company's own site.