Luel Builds Rights-Cleared Multimodal Training Data Platform for Frontier AI Teams

Training data has quietly become one of the most contested assets in artificial intelligence. Frontier labs need staggering volumes of high-quality, diverse, and – increasingly – legally defensible data to push their models forward, yet the supply side of that equation is a mess of scraped corpora, unclear licensing, and reactive lawsuits. Luel, a Y Combinator Winter 2026 startup, is building the AI training data licensing platform designed to solve that problem. The company turns everyday words and actions into usable training data, offering frontier AI teams rights-cleared multimodal datasets at scale — both bespoke collections built for specific capabilities and off-the-shelf datasets that can be dropped into training pipelines immediately.

The timing is pointed. Model developers are under growing pressure from rights-holders, regulators, and customers to prove that the data underpinning their systems is cleanly licensed. At the same time, the next generation of multimodal models – systems that fuse text, audio, images, and structured interactions – requires orders of magnitude more data than the text-only predecessors that dominated earlier cycles. Luel is positioning its AI training data licensing platform at the center of that collision – a trusted intermediary between the supply of real human expression and the demand from teams building foundation models. By handling licensing, consent, and quality control in one place, the company aims to let AI teams focus on modeling rather than data procurement.

Luel was founded in 2025 by William Namgyal and Inigo Lenderking, both UC Berkeley dropouts. Namgyal, a former student in Berkeley’s Management, Entrepreneurship, and Technology program, serves as co-founder and CEO, while Lenderking is co-founder and COO. The pair has kept the team lean – just two people – while building out a platform designed to serve some of the most demanding customers in AI. Their bet is that rights-cleared data will move from a nice-to-have to a prerequisite for serious model training, and that the infrastructure to broker that data at scale will be one of the most valuable picks-and-shovels plays of the current AI cycle.

For frontier labs, the pitch is compelling: rather than building proprietary data operations from scratch or leaning on legally ambiguous scraped corpora, they can tap Luel for curated, licensed, multimodal datasets tailored to the capabilities they are trying to unlock. For Luel, the opportunity is to become the default marketplace where the world’s training data gets cleared, packaged, and delivered – a critical layer in the AI stack that has so far been dominated by ad hoc deals and legacy vendors. Winter 2026 will be a telling batch for the data layer, and Luel is one of its more focused bets.