Designing and deploying ML models for mobile phones at scale (1M+ devices). Optimising inference performance and accelerator usage to deliver high performance on a large variety of chips.
LLM powered recruiter - Optimised query speed by redesigning schema and query logic of Elasticsearch database and improved the ranking quality by making various changes to the embedding and retrieval logic.
Conversational LLM - Created a web application and a Slack bot allowing users to interact with a company specific LLM. This model had access to company resources using retrieval augmented generation techniques.
ETL pipeline - Created a large scale distributed ETL pipeline capable of transforming, embedding and indexing 2 trillion data points in less than 48 hours.
Worked on the conceptualisation, design and build of Futon, an in-person social platform. Built the first version of the app in React Native and released this for both iOS and Android.
Responsible for developing and applying machine learning algorithms in the core product to identify forgeries, verify user liveness, and match facial biometrics to securely verify the user’s identity. My work was focused on out-of-distribution detection, unsupervised modelling and using Large Language Models to understand documents. This work led to 3 patents and core parts of the technology being used to verify over 300.000 people a day.
Co-founded a company focused on the integration of Machine Learning in existing business processes. Clients included the Dutch government, a multinational bank and various startups in the health and consumer space. I led the company's research and development of algorithms and oversaw the transition of prototypes to production models in close collaboration with the developers.