Jesse Clark

Jesse Clark

Researcher and co-founder of Celeris and Marqo .

Difference map

Celeris

Fast language and decision models. Research on parallel generation, model conversion and post-training.

celeris.ai ↗

Marqo

Multimodal search, representation learning and retrieval. Research, open models and search infrastructure.

marqo.ai ↗

Open embedding models

FashionCLIP & FashionSigLIP

22% and 57% higher average text-to-image Recall@1 than FashionCLIP2.0, respectively, across six fashion datasets.

Release benchmark
Ecommerce embeddings

Ecommerce-L improved nDCG@10 by 20.5% on average over SigLIP SO400M across the three Marqo-Ecommerce-Hard tasks.

Release benchmark

Other research & projects

Reinforcement learning

Deep Q-learning for warehouse routing at Stitch Fix, including simulations with obstacles and multiple pickers.

Warehouse mapping

Reconstructing warehouse layouts from scan timings at Stitch Fix, using matrix completion and multidimensional scaling to support routing, batching and simulation.

Physics & imaging

Coherent diffraction imaging, phase retrieval and inverse problems, including nanocrystal dynamics, dislocations and battery materials.

Writing & conversations

Wild Hearts, with Blackbird

A conversation about models, inference and building Celeris.

Listen to episode

Universal models at Amazon Robotics

Coverage of the work on shared models for robot perception.

Read coverage