Datoric CTO Jeff Lin reported that the company doubled the number of egocentric hours collected per day from 3,000 to 6,000 within two weeks, citing early investment in recruiting, training, hardware logistics, and QA pipelines, and contributor satisfaction as the two scaling lessons, with 10,000 hours as the next target.
Datoric CTO Jeff Lin announced that internal systems for problems hit while producing training data at scale have grown into Datoric Research, working on multilingual speech and paralinguistics, automated annotation and QA for egocentric video, fraud detection, data diversity, and large-scale video search, with a new Head of Research joining.
Datoric, a YC W26 company formerly known as Arzule, sells private custom training-data pipelines to frontier labs that have exhausted public data. It has a vetted pool of more than 300,000 contributors and lists data offerings spanning voice, residential video, audio-visual material, and computer-use agent samples. The company reached nearly seven figures in monthly revenue within 30 days of its YC launch, and the article says its operational moat is a contributor network that would take years to replicate.
Datoric's founder essay argues the frontier of AI data is a research problem: LLM progress has clarified what the industry understands about data needs, and closing the gap for robotics and voice requires research, experimentation, and close collaboration between model teams and the people building the collection systems.