The Idea:
In 2019, William Namgyal and Inigo Lenderking were a Fortnite duo who had never met in person. William played from the Bay Area. Inigo played from Europe. They won their first cash cup together and spent their early teens earning real money online from their bedrooms.
Years later they ended up as roommates at Berkeley. By then William had already built ezML, a computer vision startup, and kept running into the same problem. Models broke, and when he traced the failures back, the cause was almost always bad data upstream. Inigo, a machine learning researcher, was studying something adjacent: how human behaviour can be modelled and steered.
Together they saw the bigger gap. Frontier labs had scraped the internet dry. The next wave of models for robots, voice agents and wearables needed data that simply didn't exist online: first-person video of a craftsman carving a gemstone, Urdu spoken on a noisy Pakistani street, a doctor's consultation recorded in German. Someone had to pay real people to create it, with the consent paperwork to prove it.
They dropped out and joined Y Combinator. This was the beginning of Luel.
The Execution:
- Late 2025: After two years thinking about the data space, Namgyal and Lenderking left Berkeley to start Luel as part of Y Combinator's Winter 2026 batch.
- February 12, 2026: Luel launched publicly as two products in one: a catalogue of ready-to-license datasets and a collection engine that builds custom datasets to a lab's spec. The opening catalogue ran from patient-doctor conversations in South Asia to gemstone-manufacturing footage for robotics.
- May 15, 2026: Luel raised a $31.2M seed co-led by General Catalyst and Lightspeed, one of the largest seed rounds in YC history. Paul Graham joined the round, and the founders, aged 18 and 19, turned down three acquisition offers along the way. Lightspeed put the network at 500,000+ contributors across 96 countries.
- May 15, 2026 (same day): Then the internet turned on them. Avi Patel, founder of Kled, which had raised $5.5M for its own "human data marketplace", publicly accused Luel of ripping off his company. His thread hit 7.7M impressions. A follow-up post went further, alleging Luel had copied Kled's website and inflated its user numbers. Luel stayed mostly quiet.
- August 8, 2026: Instead of a rebuttal, Luel shipped. It launched Luel Labs, a research arm sitting on a network that had grown to 850,000+ contributors, up 70% in under three months. First up: the first commercial-grade text-to-speech for Tibetan and Dzongkha, two languages the major platforms don't serve at all.
- August 29, 2026: Luel Speech went live with eight voices across Tibetan, Darija, Masri and Dzongkha. On Luel's own Egyptian Arabic benchmark, its Masri voice posted the lowest error rate of every system tested, ahead of ElevenLabs and Azure.
- September 17, 2026: Luel made its speech datasets licensable over MCP, so buyers and sellers can trade data through their AI agents rather than a sales call.
- Today: Less than eight months after launch, Luel now bills itself as a multimodal lab building foundation models for machine perception, backed by $31.2M and a paid network of 850,000+ people. The marketplace was the wedge. The lab is the bet.
The lesson? When Kled's copycat accusations went viral, Luel had two options: fight it out on X or keep building. It said almost nothing and added roughly 350,000 contributors in under three months. In a market where every founder can see the same data wall, the idea was never going to be the moat. The supply side was. You can clone a website in a weekend. You can't clone 850,000 people in 96 countries.