Poindexter Labs

Poindexter Labs is built around a quiet bottleneck in frontier AI: expert data is not the same thing as annotation.\n\nThe London startup raised a £2M seed round led by Episode 1 Ventures after bootstrapping to £1.6M in revenue, according to TFN. https://t.co/P5OP81jRrE says Poindexter Labs builds expert-generated training data for advanced AI models, with workflows for complex reasoning and domain-specific knowledge.\n\nThe wedge is workflow quality. Simple annotation tools were built for boxes, labels, and short text tasks. Reasoning data needs researchers, mathematicians, lawyers, doctors, and other specialists to create and review work collaboratively, with peer-review style loops instead of adversarial pass/fail queues.\n\nThat matters as model labs push into harder reasoning. If 40-90% of generated expert tasks get discarded, better workflow is not ops polish; it changes data yield and model-training economics.\n\nQuick facts������\n● founders: Jocelyn D'Arcy\n● total capital raised: £2M seed\n● HQ: London, United Kingdom\n● Investors: Episode 1 Ventures\n\nThe larger question is whether data vendors become research-workflow companies. For frontier AI, the next scarce input may be verified expert reasoning, not more generic tokens.