Lychee Labs is an Oxford-born deeptech industrial AI lab building physics-aware industrial intelligence for mission-critical manufacturing — starting with battery manufacturing, where defect signals arrive weeks late, scrap compounds at scale, and the cost of a missed lot dwarfs the cost of catching a thousand.
Our team, advisors, and network come from Oxford, Tesla, Google DeepMind, AWS, Cambridge, MIT, Stanford, TUM, and Volta — building physics-informed hybrid AI that bakes battery-specific failure pathways into the architecture. The kind of domain priors generic ML libraries don’t carry, built by engineers who’ve operated inside the world’s most-deployed gigafactory program.
Lychee’s architecture, proprietary battery-process knowledge, and deployment history compounds with each plant.
Built by Oxford AI engineers and ex-Tesla battery operators. Anchored in peer-reviewed work: Linda Hong Cheng’s ICLR P-AGI paper on industrial dynamics foundation models, the Chronos paper from AWS — led by one of Lychee’s senior advisors — and the public battery benchmarks Lychee validates across, including the canonical MIT-Stanford-SLAC dataset, Oxford’s NMC pouch benchmark, and the BatteryLife Na-ion dataset (HKUST-GZ). Six public datasets, five labs, four chemistries (LFP, LCO, NMC, Na-ion). Advised by senior leaders from Google DeepMind, AWS Chronos, and the World Energy Council.
Battery economics are extreme — gigafactory ramps lose ~€10M per yield point at 40 GWh, mineral efficiency is supply-chain-critical under EU Battery Regulation, and the regulatory clock is ticking. Lychee’s architecture generalizes wherever defect signals arrive too late and the cost of being wrong is high — energy, pharma, semiconductor, aerospace, nuclear, defense.
Born in Oxford. Built for the world.
