Encord leverages financial micro-models for data annotation

After meeting at an entrepreneur matchmaking event, Ulrik Hansen and Eric Landau teamed up to leverage their experience in financial trading systems into a platform for faster data labeling .

In 2020, the finance industry veteran couple founded Encord to adapt micromodels typical of finance to automated data annotation. Micromodels are neural networks that require less time to deploy because they are trained on less data and used for specific tasks.

Encord’s NVIDIA GPU-based service promises to automate up to 99% of enterprises’ manual data labeling with its microtemplates.

“Instead of building one big model that does everything, we just combine a lot of smaller models together, and that’s very similar to how a lot of these commercial systems work,” Landau said.

The London-based startup recently landed $12.5 million in Series A funding.

Encord is an NVIDIA Metropolis Partner and member of NVIDIA Inception, a program that provides go-to-market support, expertise, and technology for AI, data science, and HPC startups. NVIDIA Metropolis is an application framework that makes it easier for developers to combine video cameras and sensors with AI-enabled video analytics.

The company said it has attracted companies in the fields of gastrointestinal endoscopy, radiology, thermal imaging, smart cities, agriculture, autonomous transportation and retail applications.

“Augment Physicians” for SurgEase

In 2021, the partners hunkered down near Laguna Beach, California, at the home of Landau’s parents, to build Encord while attending Y Combinator. And they had also just landed their first customer, SurgEase.

London-based SurgEase offers telepresence technology for gastroenterology. The company’s hardware device and software allow remote physicians to monitor high-definition images and video captured during colonoscopies.

“You could have a doctor in an emerging economy to do the diagnosis or the detection, as well as a doctor from one of the best hospitals in the United States,” Hansen said.

To improve diagnostics, SurgEase also applies video data to train AI models for detection. Encord microtemplates are applied to annotate video data used for SurgEase templates. The idea is to give doctors a second look at the procedures.

“Encord’s software has helped us solve some of the toughest problems in endoscopic disease assessment,” said Fareed Iqbal, CEO of SurgEase.

With AI-assisted diagnostics, clinicians using SurgEase could catch more stuff sooner so people don’t need harsher procedures down the line, Hansen said. Doctors don’t always agree either, so it may help reduce noise with another opinion, Landau said.

“It’s really increasing doctors,” Landau said.

King’s College London: 6 times faster

King’s College London was challenged to find a way to annotate images in videos of precancerous polyps. So she turned to Encord for annotation automation, because using highly trained clinicians was expensive on such large datasets.

The result was that micro-templates could be used to annotate approximately 6.4 times faster than manual labeling. It was able to handle about 97% of datasets with automated annotation, with the rest requiring manual labeling by clinicians.

Encord has enabled King’s College London to reduce model development time from one year to two months, bringing AI to production faster.

Triton: quickly in inference

Encord initially planned to build its own inference engine, running on its API server. But Hansen and Landau decided that using NVIDIA Triton would save a lot of engineering time and get them into production quickly.

Triton offers open source software to put AI into production by making it easy to run models in any framework and on any GPU or CPU for all types of inference.

Additionally, it allowed them to focus on their first customers by not having to architect the inference engine themselves.

People using Encord’s platform can train a micromodel and run inference very soon after, thanks to Triton, Hansen said.

“With Triton, we get native support for all these machine learning libraries like PyTorch and it’s compatible with CUDA,” Hansen said. “It saved us a lot of time and hassle.”

Comments are closed.