About
Co-founder @marqo.ai. Building the next generation of search.
Contributor podcast - “Vector Search for Humans - Marqo with Jesse Clark”
Multimodal vector search with personalization
Open CLIP benchmarks
GPT & LLM integration/augmented retrieval/retrieval augmented generation
Image search with localization
Searching over stable diffusion generated images
Marqo and tensor search
Background
Prior to Marqo, I was a machine learning (ML) lead working at Amazon Robotics AI (RAI) within the ML and computer vision team. I designed and led the implementation of Janus a continual learning framework. I also led the development of the deep learning models for the perception system of Amazon’s Robin robot. Prior to RAI, I was leading the development of ML for new technologies, working at the intersection of the smart home and IoT. Before Amazon, I worked at Stitch Fix doing a mix of data science, machine learning and operations research. Prior to industry, I was a physicist working to develop lensless imaging techniques by solving inverse problems using non-convex optimization methods. I was the recipient of a Volkswagen Foundation Paul-Ewald Fellowship while working at Stanford and prior to that I was working at University College London. This research resulted in a number of significant achievements in nanoscale imaging and high impact publications in journals such as Science, Nature Physics, Nature Materials, Nature Communications, PNAS, and Physical Review Letters (see Google scholar link to publications list).
See below for more links to work and projects.
Amazons Robin robot
Training neural networks with iterative projection algorithms
Using projection based optimization to train neural networks Link.
The curious connection between warehouse maps, movie recommendations, and structural biology
Exploring how we are combining ideas from recommender systems and structural biology to automatically draw warehouse layouts and track when they change Link.
Be smarter. Be seetd.
A data driven and optimization centric approach to allocating seats and desks to people in an office Link.
Skynet Salesman
Using reinforcement learning to solve routing problems Link.
Sudoku solver
Using projection methods to solve Sudoku Link.
YouTube
A couple of movies from academic projects Link.
Academic Projects (selected)
“Ultrafast Three-Dimensional Imaging of Lattice Dynamics in Individual Gold Nanocrystals” Science - seminal work demonstrating the imaging of phonons in nanocrystals (Link). Also featured as a “Perspective” (Link).
“Three-dimensional imaging of dislocation propagation during crystal growth and dissolution” Nature Materials - demonstrating the imaging of dislocations within nano-crystals as they grow (Link). Also highlighted for “News and views” (Link). This work also allowed imaging of defects within battery nano-particles and was subsequently published in Science (Link).
“Imaging transient melting of a nanocrystal using an X-ray laser” PNAS - a continuation of the work on developing ultra-fast imaging of nanomaterials (Link). Also highlighted in Nature Physics research highlights (Link).
A full list of academic papers can be found in the Google scholar link.