Singapore
Machine learning engineer with experience in reinforcement learning and natural language processing. Masters of Data Science & Machine Learning graduate from the National University of Singapore. Previously graduated with a BEng. Mechanical Engineering, with a specialisation in robotics.
• Developed and deployed a multitask pairwise CTR ranking model (MMoE + ESMM), increasing LIVE notice CTR 13.18% → 15.45% (+17.3%), expanding reach +12.9%, and reducing effective CPC by 93%. • Built a scalable multilingual text retrieval and multi-label classification pipeline for livestream transcripts, finetuning SLMs for multi-label classification with attention-based pooling and label-aware heads to achieve 94.5% precision and 86.1% recall across core content elements. • Implemented configurable LLM-based content understanding workflows with structured prompting, optimizing the inference funnel to eliminate redundant compute and reduce pipeline cost by ~60%. • Finetuned and aligned LLMs for high-quality livestream clip classification using SFT + RL, increasing baseline precision from 39% to 76% and human acceptance rate from 35.9% to 52.8% (+47%). • Designed automated visual content curation and LIVE creator persona pipelines (31 personas, 24 regions), achieving 98%+ human acceptance and ~90% detection accuracy. • Scaled online inference APIs and offline pipelines to up to 1B+ users, improving throughput 2.9×, cutting runtime 50%, and reducing memory usage 52%.
• Designed & delivered a reinforcement learning (RL) product enabling no-code RL training & evaluation workflows • Implemented RL algorithms and custom PyTorch model components for effective multi-agent RL • Developed end-to-end retrieval augmented generation (RAG) solutions for LLM applications, including multimodal PDF extraction, retrieval / reranking and document management / data annotation workflows
• Established a weak-supervision workflow to train machine learning models with limited labelled data • Conducted text clustering and topic modelling with BERTopic to decipher user and content patterns • Worked with stakeholders to support ad-hoc data analytics for minimising platform content risk
• Optimised paid ads promotions to improve ads revenue uplift by 32% & program ROI by 13% • Built dashboards & automation pipelines with SQL, Python (Pandas) & Excel • Conducted root-cause analysis on business metrics to inform business strategy
[A*STAR Research Internship Award] Robot Active Learning with Human-in-the-loop Reinforcement Learning • Researched Human-in-the-loop Reinforcement Learning (HRL) & Imitation Learning techniques • Designed & implemented sample-efficient HRL methodologies in PyTorch • Built a human-in-the-loop interface for reinforcement learning with PySimpleGUI