Tech Lead of Foundation Models at Snorkel AI, San Francisco.
I specialize in Artificial Intelligence and Software with 7 years of experience and a Masters degree from the University of Cambridge in Information and Computer Engineering.
Please feel free to reach out. I am always open to new opportunities, particularly those in Machine Learning. I look forward to hearing from you!
Scope is bursting the information bubble by exposing diversity of opinion on the breaking news headlines through AI. It is a news platform (Android, IOS & Web) that aggregates multimedia breaking news content from across the globe. It utilizes machine learning to group all of the content about the same headline together. Providing you with a diversity of opinion to help you combat fake news and bias.
Android IOS Web
Contracted by CERN to commercially develop our project from a hackathon held at their Geneva headquarters. The system consists of an intelligent matching engine which links people with projects using natural language processing techniques, machine learning, and an intuitive user interface.
In 24 hours we developed an automated on-street parking management system. The project involved hacking microcontrollers with magnetometers to detect cars. The devices relayed information to the internet where a web interface displayed in real-time available on-street parking spaces. We placed second in the competition.
A multi rotor UAV controlled via `First Person View` from a smartphone within a VR headset. A live feed from a camera placed on the drone is relayed to the headset and the drone can be controlled by simple head turns. I led a small team to build this from scratch covering the mechanical, electrical and software design and build. We secured £500 funding from the University of Cambridge Engineering Department.
GitHub
An autonomous vehicle that collects, identifies and then delivers different objects to their corresponding trays. I was lead software engineer of the six person team; writing algorithms for line following, identification and control.
Snorkel AI is the data centric AI company for enterprize, making model development look and feel like software development.
I have been developing everything from model benchmarking, machine learning based ranking, to developing Snorkel's Foundation Model capabilities from the ground up. I am now the tech lead for all of Snorkel's Foundation Model capabilities, a highly cross-functional area of the upmost importance to the company and our clients.
Eigen is an AI company developing state-of-the-art NLP algorithms to extract insights from client's data; their ML models only require 2-50 examples to train and achieve industry leading accuracy.
· Developed two cascaded models. Compared to Eigen’s old production model, one was more lightweight (10x faster and 6x less memory) and the second more accurate (2x faster, 2x less memory).
· Created a deep learning based webapp for table cell extraction, deployed via docker microservices and used in parallel across four departments.
· Implemented a Graph Neural Network for the inferring structure of tables in documents, utilizing Monte Carlo sampling over cliques to reduce memory consumption when training the model.
· Implemented a Probabilistic Graphical Model for table cell prediction; F1 scores of 90%+ across a variety of tasks.
· Designed and delivered a Deep Learning lecture series to the research team consisting of 6 presentations.
Scope is bursting the filter bubble by exposing diversity of opinion on the breaking news headlines through AI.
· Accepted onto competitive Cambridge Judge Business School Accelerator Programme.
· Won place at finale of national competition Pitch@Palace. Pitched to senior executives and royalty at St James’s Palace.
· Acquired client for news data service with two more in the pipeline.
· Interviewed candidates and led small team of three people.
· Developed full stack; algorithms -> databases -> server -> client backend -> client frontend
· Implemented an Automated Deployment and Testing pipeline inline with Agile Development Methodologies.
· Parallelized work using multi-threading through asynchronous functions.
· Developed both native mobile applications (Android, IOS) and a progressive web application.
· Designed and trained both supervised and unsupervised NLP algorithms for Sentiment Analysis, News Clustering, Topic Classification and Summarization.
· Developed a word embedding using TFiDF; trained a cutting edge CNN, RNN and Random Forest Classifier.
· Deployed scalable algorithms to the cloud using Amazon Web Services running in real time.
Consulted for a high tech company in the medical devices space and a legal tech startup.
· Developed bespoke Automated Deployment script utilizing Docker for deploying to AWS’s serverless architecture, Lambda.
· Integrated cloud deployment into the IOS application along with automated push notifications.
· Utilized Agile Development methodologies and coded using OOP practices.
· Developed full-stack for a responsive web app including features such as the matching algorithm, payments and in-site messaging.
· Developed extensive pre-processing pipeline rendering a noisy, difficult dataset tractable.
· Researched and led the development of an RNN classifier to detect heart arrhythmias achieving accuracy of over 95%.
· Deployed Deep Learning algorithm to the cloud using cutting edge AWS serverless technology.
The Technology Partnership plc (TTP) is Europe's leading technology and product development company.
· Only intern to start a project and have it subsequently sold to a client.
· Offered full time position.
· Developed algorithm and system to remotely detect human activity.
· Optimized methods in memory and power consumption for a severely constrained device.
· Used Linux in both development and for production environment.
· Researched and developed a CNN in TensorFlow to identify cars from infra-red images.
· Utilized Transfer Learning and novel dataset generation methods to train a CNN with no pre-existing dataset.
Cubica build cutting edge algorithms, software, and systems for processing images & video.
· Developed on a Linux based system in Python.
· Integrated ML solution into their module containerized solution using Docker.
· Implemented a deep learning paper on object localization and classification within images. This significantly improved the speed of their flagship computer vision product.
· Designed and trained a sequence to sequence model for speech to text classification in TensorFlow.
BAE Systems plc is the UK's largest defense, security, and aerospace company.
· Aided in producing the campaign plan to try and secure contracts from the MoD totalling £50 Million.
· The campaign plan included strategy, SWOT analysis, stakeholder analysis & costing.
· Designed a solution to accurately decode GPS data into appropriate messages using UML framework and OOP techniques.
· Completed the solution to a professional standard for it to be tested on the next submarine sea trials.
Accelerate Cambridge selects the most promising early stage startups in the Cambridge area every year. It helps the entrepreneurs turn an idea into a business, launch, and then keep going. Accelerate Cambridge is part of an ecosystem of support for entrepreneurs in Cambridge, helping them navigate their way through it and find the resources they need.
Awarded 2.i MEng BA with Distinction in Masters Project.
Notable Courses: Probabilistic Machine Learning, Deep Learning and Structured Data, Statistical Signal Analysis, Software Engineering and Design, Practical Optimisation, Bayesian Inference.
Masters Thesis: Project aimed to help reduce the negative bias of news through deep learning sentiment analysis. Created a novel architecture combining a CNN and RNN yielding good performance.
The Advance Leadership Award aims to provide support for engineering undergraduates in UK universities who have the potential to become leaders in engineering and who are able to act as role models for future engineers. These awards help ambitious and inspiring engineering undergraduates to undertake an accelerated personal development programme.
In 24 hours, we developed an automated on-street parking management system. The project involved hacking microcontrollers with magnetometers to detect cars. These devices then communicate with each other via Bluetooth low-energy to pass a signal to the internet via a LoRa network. The information is presented in real-time via a web interface to enable real-time detection of on-street parking spaces. We placed second in the competition.
Cambridge University Entrepreneurs run an annual Young Entrepreneur of the Year competition. Approximately 200 people enter the competition and after I reached the final stage of interviews consisting of under 10 candidates. I won the best pitch prize.
The Defence Technical Undergraduate Scheme provides leadership training and character building through the means of high stress, physically and mentally demanding situations. These are predominantly out in the field and all have the aim of developing individuals into future military leaders.
