Poindexter Labs secures £1.9m seed round for AI reasoning datasets
Published
Olympiad-grade STEM datasets and stepwise reasoning annotations are the core products of Poindexter Labs, which has just secured an estimated £1.9m seed round. Based in London, the company specializes in creating high-level data used to train and test frontier AI models.
Incorporated in August 2025, Poindexter Labs provides materials for AI research labs, including multi-step reasoning tasks, proofs, and chain-of-thought problems. These are delivered in LaTeX and structured JSON or Parquet formats to support reinforcement learning and benchmark development. The company currently employs 33 people.
The firm's work involves decomposing solutions into verifiable steps and tagging error types or alternate paths. This data supports instruction-tuning and the creation of reasoning curricula. Poindexter Labs also designs benchmarks and leaderboards that include leakage tests and reproducible scoring.
To maintain the quality of its data, the company uses a vetting system for its contributor pools. This includes secure assessments to evaluate math and logic ability, using techniques such as hidden gold and cross-item invariants to detect AI-generated text or tool-overreliance. The company also performs red-teaming to identify failure taxonomies and hallucinations in math and logic models.
Recent corporate filings from April show the appointment of Dr Shuaib Adam as a director. LinkedIn data suggests Dr Adam is possibly a General Partner at Episode 1 Ventures, though this association is unverified. The documentation for the £1.9m seed round was processed in April.
Contact the editorial team at [email protected]