Leap Labs

Leap Labs (Leap Laboratories Inc.) is an AI interpretability startup founded in early 2023 by Jessica Rumbelow, who holds a PhD in machine learning with a focus on model-agnostic interpretability. The company is incorporated in the US as a for-profit entity and operates primarily from London, UK. Leap Labs began with an explicit AI safety mission: building a universal interpretability engine that could take any neural network and return human-parseable explanations of what the model has learned. Their research philosophy emphasizes reproducibility, generalizability, and theoretically motivated methods rather than ad-hoc heuristics, and is designed to be model-agnostic so that it remains useful as AI architectures evolve. Over time, the company has focused its interpretability capabilities on scientific discovery. Their Discovery Engine platform fits neural networks to tabular datasets and then applies proprietary interpretability algorithms to extract meaningful patterns, validating findings on held-out data and contextualizing them with existing literature. The platform delivers ranked, statistically significant discoveries with p-values and effect sizes, and is accessible via a web interface, Python SDK, REST API, and MCP server. The company has documented over 1,300 novel findings across more than 600 datasets, with early results spanning plant biology, meteorology, immunology, and materials science. Notable discoveries include a novel genotype-nutrient combination improving crop resilience and evidence that disproves a foundational assumption in meteorology with implications for climate modeling. The team includes Jessica Rumbelow (Founder & CEO), Jugal Patel (Founder & COO), Robbie McCorkell (CTO), Zohreh Shams (CSO, Cambridge Industrial Fellow and AI PhD), and several ML engineers and researchers. Leap Labs won the Novel AI track at the Deep Tech Momentum DTM100 pitch competition and has received a grant from Open Philanthropy (via Coefficient Giving's Navigating Transformative AI fund) for interpretability research.