Analytics on PayLater service

• Integrate Role-Based Access Control into Temasek’s data search engine• Implement health checks for Temasek’s data search service• Build Celery service to offload Elastic Search tasks from main server to ensure smooth user requests stream on the frontend • Develop model baselining framework to evaluate results from different NLP models used in Temasek’s search engine - Decision support and Rank aware metrics

Video Analytics team in Data Science and Artificial Intelligence Division (DSAID)• Build data pipeline to ingest video frames and pre-process into individual frames as model input• Independently implemented a CNN model to perform multi-class classification on different breeds of mosquitoes • Build upon OpenCV pretrained models to detect misalignment in shoulder frames on portrait photographs

Training Convolutional Neural Network for ARMA Order predictionWrote scripts in Python for: • Generating synthetic stable and invertible time series data for training• Building CNN with skip connections in PyTorch • Model training based on generated training dataReferences:1) Uniform Random Parameter Generation of Stable Minimum-Phase Real ARMA(p, q) Processes2) Numerical Recipes in C: Root of Polynomials2) Bayesian Statistics Without Tears: A Sampling-Resampling Perspective

• Developed sequence model with LSTM/GRU layers and Multi-head Attention mechanism using PyTorch for text augmentation• Topic Modelling with LDA, K-Means, Agglomerative Clustering to discover content blocks within chat bot history• Performance comparison between BERT and RoBERTa for text classification• Customer segmentation analysis for upcoming e-commerce platform

• Apply Natural Language Processing (NLP) for text summarization on legal documents for improved readability • Perform extractive and abstractive summarization on lengthy text documents; methods used includes word embedding, auto encoders, clustering• Models explored - Google BERT, Microsoft UniLM• Hyperparameter optimisation in Deep Learning model to achieve highly accurate summaries• Full stack web development with Django, React.js and SQLite for a project showcased in IoT Asia+

• Web development with Flask and MS SQL for Hyundai warehouse to track employees’ attendance and overtime hours• Statistical modelling on data provided by Hyundai warehouse to track shipment flow and analyse business flaws; methods used includes k-means, hierarchical clustering and various convex optimisation algorithms• Data management and preprocessing in MongoDB and R respectively