cv

PhD Student

Simula Research Laboratory

• Researching causal representation learning as means of improving robustness and generalisation capabilities of Neural Network models
• Designing and building novel causal disentanglement models as well as implementing and testing existing state-of-the-art models in the field

AI Researcher

Firi AS

• Built, trained, and tested a variety of NN architectures, including bespoke Transformers and Attention-based Graph Neural Networks for predictive risk modelling
• Designed and implemented from scratch a microservice backend architecture to underpin efficient and scalable data extraction, processing, and feature engineering

Co-Founder, CTO

AIUS

• Designed and implemented ML-based document and sequence classification models, and pitched these to our non-technical customers
• Built back-end API and database schemas for our document management software, using Apollo GraphQL and MongoDB

Data Science Intern

Techspert

• Developed a collection of multilabel classifiers using Tensorflow / Keras, for categorising and filtering incoming company emails
• Built pre-processing pipeline to clean, anonymise, and build feature vectors from training data

Undergraduate Researcher

University of Cambridge

• Developed and modified lattice-based RNN models for improved speech recognition confidence scores
• Designed and built automated system for grade-appropriate feedback to English learners. Later released as part of the app "Sandi"

Co-Founder, Vice President

Cambridge University 3D-Printing Society

Project leader in collaboration with Cambridge-based charity REMAP to upgrade remote-controlled archery turret for users with movement disabilities

M.Eng.

University of Cambridge

Information and Computer Engineering

• Probabilistic Machine Learning
• Deep Learning
• Computer Vision

King's Scholar

King's College Cambridge

Received in 2016 and 2017 for excellent exam results

Measuring the Effect of Causal Disentanglement on the Adversarial Robustness of Neural Network Models

Proceedings of the 32nd ACM International Conference on Information and Knowledge Management

Investigated relationship between causal disentanglement and robustness in state-of-the-art models

Increasing context for estimating confidence scores in automatic speech recognition

IEEE/ACM Transactions on Audio, Speech, and Language Processing

Suggested, implemented and wrote up novel approach for Confusion Network to sequence aligment using a soft Levenshtein loss

Bi-directional Lattice Recurrent Neural Networks for Confidence Estimation

IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

Conducted experiments (TensorFlow and PyTorch) and co-authored research paper, in conjunction with final-year academic commitments