Celeris is an artificial intelligence research lab building the world's fastest language models. We are creating a future where frontier intelligence responds in microseconds.
Real-time voice systems, interactive coding environments, search, autonomous agents, robotics, and scientific infrastructure all depend on AI that can respond within milliseconds or microseconds. Today, builders of these systems must settle for smaller, weaker models, or ration frontier intelligence because it cannot keep up. We believe the next generation of AI will not be defined solely by intelligence. It will be defined by intelligence that operates in real time.
Rethinking inference
Rather than treating latency as an engineering optimization applied after training, we treat latency as a fundamental research problem. Our work explores new algorithmic methods that improve the speed of language generation and the inference systems required to serve them, while preserving frontier-level capability. Our objective is simple: maximize useful intelligence delivered per unit of time.
Looking forward
The history of computing is a history of reducing latency. From networking to databases to GPUs, every major improvement in responsiveness has expanded what software could become. AI is entering the same transition. As models become instantaneous, they move from tools that generate responses to systems that participate continuously alongside people and software. The last decade of AI was defined by scaling capability. The next will be defined by scaling speed, and we are building the models for that future.
Our team
We are researchers, engineers, and entrepreneurs who have spent our careers building AI systems for production. Our team and investors have held technical and leadership roles at organizations including Amazon, OpenAI, Google Brain, Cohere, Stanford, University College London, Cambridge, and leading AI startups, contributing to advances in foundation models, inference systems, and machine learning infrastructure used by millions worldwide.
Celeris was founded by Tom Hamer and Jesse Clark, who previously founded Marqo, whose search infrastructure now powers billions of dollars of e-commerce. Tom built core infrastructure as an engineer at AWS and holds a machine learning master's from Cambridge. Jesse led machine learning at Amazon Robotics and was principal ML scientist at Stitch Fix, following physics research posts at Stanford and UCL. Celeris takes on a different problem - inference speed rather than retrieval - but it is the same ambition: solving hard, foundational infrastructure problems in AI.
