Apr 2023 – Aug 2024 · 1 yr 4 mos

• Google Deepmind
• Training the most capable multimodal embedding models
• Improving vision encoders by leveraging self-supervision along with language supervision
• Large scale vision-language models e.g. Multimodal MUM (https://blog.google/products/search/introducing-mum/) and contrastive representation learning (https://arxiv.org/abs/2212.04979

Jun 2019 – Apr 2023 · 3 yrs 10 mos

• Cloud AI Video Intelligence:
• Action recognition and localization
• Video search

Dec 2018 – May 2019 · 5 mos

• Working on deep neural networks for Simultaneous Localization and Mapping
• Improving sample-efficiency of meta-reinforcement learning through concurrent rollouts and reward shaping.
• Advisor: Ruslan Salakhutdinov

Aug 2018 – Nov 2018 · 3 mos

• TA for 11-785 Introduction to Deep Learning (Fall 2018) under Prof Bhiksha Raj
• Designing homework assignments
• Holding office hours
• Conducted recitations on CNNs and VAE
• Mentoring deep learning project teams.
• This course is primarily conducted in PyTorch.

Jan 2018 – May 2018 · 4 mos

• TA For 10-601 Introduction to Machine Learning (Masters) Spring 2018 under Prof Matt Gormley
• Conducted recitations
• Regular office hours
• Setting exam questions and grading
• Specifically focused on MLP, MAP, Naive Bayes, decision trees part of syllabus.

May 2018 – Aug 2018 · 3 mos · Greater Seattle Area

• Implemented neural early fusion and double fusion semantic similarity models in TensorFlow for word-sense disambiguation given context in home automation using voice assistants
• Used PySpark to process and extract large stream datasets and perform feature processing

May 2016 – Aug 2016 · 3 mos · Singapore

• Proposed denoising auto-encoder architectures for traffic speed and travel time estimation large urban road networks from sparse raw GPS sensor data implemented in Theano
• Shape File data fusion from multiple data sources to enhance accuracy of nearest neighbour traffic estimation.

Feb 2016 – Jul 2017 · 1 yr 5 mos · Singapore

• Led development from scratch of web-based large-scale data driven agricultural risk pricing enterprise software with automatic data-cleaning capabilities and visualisations built using AngularJS, Node.js and PostgreSQL for primary insurance companies (Innovation award at 7th Asia Insurance Technology Awards).
• Developed a full-stack prototype desktop dashboard prototype application for managing parametric insurance products.

May 2015 – Aug 2015 · 3 mos

• Extended internal web tools to be used for tagging and untagging database entries which made the process more efficient.
• Built a dashboard using AngularJS with graphs drawn using d3.js for viewing analytics data which is part of a product.

Mar 2012 – Jun 2012 · 3 mos · Singapore

• Worked on design of website for company, and back end processing as well as front end development of an intelligent, PHP server based question bank and attendance system for students.

Jan 2012 – Mar 2012 · 2 mos · Singapore

• Conducted classes for two JC1 classes for a term, teaching H2 level Mathematics, and also handled administrative work for the Mathematics department.