- #AIforPublicGood as part of the "Computer Vision +" Tech Practice, under the Technology Associate Program.
- Cooking up baseline models, tooling, benchmarks, evaluation protocols and best practices for detection of AI-generated content and misinformation.
- Solo leveling the engineering and management of an internal "Playground", which showcases over 100 multimodal AI capabilities, use cases, and models. This initiative has empowered the team to deliver impactful pre-sales demonstrations across various roadshows, exhibitions, and conferences, significantly accelerating the conversion of leads. It has also facilitated detailed feasibility studies, greenfield and whitespace experiments, and the rapid prototyping of new concepts and initiatives.
- Translating cutting-edge research in large multimodal models (LMMs) and retrieval-augmented generation (RAG) into practical, applicable use cases tailored for public agencies. This effort effectively bridges the gap between advanced AI research and real-world AI-enabled applications for public good.
- Optimized and updated edge vision pipelines for the classification, detection, and reidentification of illegal smoking in public spaces, specifically designed to meet the unique requirements of the National Environment Agency of Singapore. The project has achieved optimal performance while maintaining real-time processing capabilities, and has dramatically reduced false positives and unwanted multiple counts by 90.17%, setting the stage for larger-scale operationalization.
Work Experience
- Improved performance of advertisement moderation model pipelines through the investigation and implementation of new training methodologies and algorithms from areas such as metric learning, semi-supervised learning, and face forgery detection, using Python and PyTorch, as part of the monetization integrity team.
- Built development tools to facilitate data cleaning, fast model iteration, and streamlined model development pipeline.
- Implemented autonomous data acquisition and annotation pipeline using depth, tracking and infrared cameras on a Raspberry Pi-mounted drone, pseudo-labelling, and semi-supervised learning, reducing data collection and annotation time by 90%.
- Trained and deployed detection, segmentation and tracking models in an iterative and incremental manner using PyTorch and TorchServe on AWS EC2 for Agri-tech specific tasks, ensuring high mean average precision and throughput.
Education
- Awarded the Lee Kuan Yew Gold Medal for being the top student in the cohort.
- Awarded Dean's List for all eligible years, ranking in top 5% of cohort with a GPA of 4.97/5.00 (Honors – Highest Distinction).