Labeling and annotation platforms may not get the eye that flashy recent generative AI models get. But they’re essential. The data that many models are trained on should be labeled, otherwise the models wouldn’t give you the chance to interpret that data during training.
Annotation is a large undertaking, requiring 1000’s to thousands and thousands of annotations for the larger and more demanding datasets. To ease the burden, Eric Landau and Ulrik Hansen founded Encordwhich they describe as a “data development platform” for firms that manage and prepare their data for AI models.
Thanks to a Series B funding round led by Next47, the corporate now has an extra $30 million in its coffers, bringing Encord's war chest to $50 million. The recent capital will probably be used over the following six months to double the scale of Encord's product, engineering and AI research teams and expand the corporate's San Francisco offices, Landau told TechCrunch.
“We aim to grow our team from the present 70 to 100 people by the top of the yr,” he added. “We now have two headquarters in London and San Francisco with team members all around the world.”
Landau began working with big data systems while at Stanford, conducting research in particle physics. Hansen worked at JP Morgan in global markets, trading emerging market derivatives.
Hansen says the thought for Encord got here to him while working on data-intensive AI projects during his computer science master's degree at Imperial College London. Frustrated by the time-consuming data curation and labeling process, Hansen met with Landau, whom he knew from the London entrepreneurial scene, to work together to seek out ways to resolve the info problem.
“By combining Hansen's expertise in software development with my insights from quantitative research to automate data development, we launched the primary iteration of Encord's product during Y Combinator in spring 2021,” Landau told TechCrunch. “The Encord platform equips firms with tools to arrange their data for AI and assess how effectively that data supports their models.”
Given the scale of the info annotation and labeling market appreciated Encord, which is anticipated to grow to $3.6 billion by 2027, is one among many vendors competing for business. Beyond the elephant within the room – Scale AI – there are startups like Datasaur, which lets customers routinely create models from sets of labels; Heartex, which is constructing an open-source platform for “engineering” data; and data annotation tools provider Dataloop.
Encord stands out due to the versatility of its platform, says Landau.
Encord enables teams to explore and visualize datasets—including image, video, and voice datasets—from private and public cloud storage, and compare the performance of various models trained on the identical datasets. The platform attempts to discover issues with model accuracy and suggest additional training data that would help address those issues.
“Unlike piecemeal solutions that only address certain parts of your data stack, Encord lets you consolidate your entire data workflows into one platform,” said Landau. “This consolidation gives firms traceability that sheds light on the customarily opaque 'black box' of AI and helps understand why a model makes certain decisions.”
Encord's strategy appears to be working well to this point. The company has 120 customers, including Philips, emerging AI startup Synthesia, and healthcare providers Cedars-Sinai and Northwell Health, in addition to contracts with undisclosed military and government agencies. Landau claims that Encord quadrupled its revenue last yr and that it might be money flow positive by 2025 if it didn't proceed so as to add headcount.
“We are experiencing the alternative of a slowdown,” said Landau. “However, we’re aware of the final market conditions and have taken a conservative approach to capital deployment.”
Other participants in the brand new financing round were Y Combinator, CRV and Crane Venture Partners.
