Research Interests
current focus
Multimodal & VLMsVision-language pretraining, visual question answering, joint embedding spaces
Language Model PretrainingMixture-of-Experts SLMs, multi-token prediction, efficient attention, RAG
Generative ModelingDiffusion models, DDPM/DDIM sampling, controllable & latent-space editing
Efficient InferenceQuantization, knowledge distillation, ONNX / TensorRT deployment
Video UnderstandingReal-time multi-object tracking, re-identification, pose & depth estimation
Retrieval & SearchVector databases, hybrid search, semantic reranking, FAISS
Selected Projects
research & applied
All
Language Models
Vision
Generative
Systems
Custom Mixture-of-Experts Small Language Model

Built and pretrained a PyTorch MoE SLM on Gated DeltaNet 2, with multi-token prediction, fill-in-the-middle, and context-length curriculum. Includes cached Parquet loading, online sequence packing, custom Triton kernels, Muon + 8-bit AdamW optimization, and WSD scheduling. Trained on 1B+ tokens (130k vocabulary) to a pretraining perplexity of ~20 with closely matched train/val curves.

PyTorchMoETritonSFTFill-in-the-Middle
Text-to-Image Diffusion — From Scratch

Conditional diffusion training framework with DDPM, mixed precision, cached text embeddings, DDIM sampling, and a custom U-Net. A second research pass added dynamic aspect-ratio batching, automated captioning, and resumable Jupyter-based training workflows.

DiffusionDDPM/DDIMU-NetPyTorch
Retrieval-Augmented Generation Chatbot

Production RAG system for internal knowledge sources using FastAPI, LangChain, and Pinecone. Implemented document ingestion, vector embeddings, semantic retrieval, prompt orchestration, and REST deployment for structured and unstructured data.

RAGLangChainPineconeFastAPI
Real-time Vehicle Monitoring Pipeline

End-to-end RTSP/IP-camera analytics: custom YOLOv4 training on AWS SageMaker, automatic license-plate recognition, multi-object tracking, and entry/exit monitoring. Achieved 45 FPS on GPU with 90% recognition accuracy after quantization.

YOLOv4ALPRSageMakerTracking
Multi-angle Face Recognition Pipeline

Real-time IP-camera face recognition using rotated face detection, embedding extraction, and approximate-nearest-neighbour search with FAISS and Annoy for robust identity matching across viewing angles, on both CPU and CUDA.

FAISSAnnoyFace Recognition
Learned Image Compression (VQGAN)

Compression pipeline built on pretrained VQGAN / KL autoencoders with custom serialization and reconstruction, producing file sizes comparable to 20–30% JPEG compression at higher perceptual quality.

VQGANAutoencodersCompression
On-device Driver Safety System

Android drowsiness and head-pose detection system using quantized TensorFlow Lite models with Java/C++ integration and OpenCV DNN for low-latency, on-device inference.

TensorFlow LiteAndroidOpenCV DNN
Paper Implementations — Deep-dive Series

From-scratch PyTorch re-implementations of landmark papers spanning GANs, diffusion, segmentation, recognition, and knowledge distillation, built to deepen architecture-level understanding and reproducibility.

ResearchPyTorchReproducibility
Software Applications
interactive demos
All
NLP
Vision
Tokenizer Visualizer

Interactive tool for visualizing how different tokenizers split and encode text, useful for inspecting vocabulary behavior across language models.

TokenizationNLP
RAG Chat Visualizer

Visualization interface for retrieval-augmented generation chat pipelines, showing retrieved context, embeddings, and response grounding.

RAGRetrieval
Image Processing NodeUI

Node-based visual editor for building and chaining image processing operations interactively, without writing code.

Image ProcessingNode UI
Experience
roles
AI Engineer — FreelanceComputer vision, LLM systems, deep learning
AI Engineer — Full-timeComputer vision, NLP, deep learning
AI Engineer — InternshipChatbot development
Undergraduate Teaching AssistantData Structures Lab · Object-Oriented Programming Lab
Undergraduate Teaching AssistantDatabase Systems Lab · Compiler Lab
Core Skills
toolkit

Computer Vision

Detection, tracking, segmentation, pose & keypoint, depth estimation, image captioning, real-time RTSP pipelines

Language & Retrieval

Custom SLM pretraining & SFT, LangChain, Hugging Face Transformers, vector databases, hybrid search, reranking

Generative AI

Diffusion models, image editing, latent control, from-scratch and fine-tuning training pipelines

Deployment & Inference

Ollama, llama.cpp, ONNX Runtime, quantization, GPU inference optimization, FastAPI

Multimodal & Speech

Vision-language models, VQA, image-text joint embedding, speech-to-text / text-to-speech pipelines

Engineering

Python, C++, Java, PyTorch, TensorFlow, OpenCV, AWS (S3, EC2, SageMaker), Pinecone, Linux

Education
degree

B.Sc. in Computer Science and Engineering

Training & Certification
selected
NUS ISS / ICT Ministry — Specialist Training Programme in Artificial Intelligence (Deep Learning, Computer Vision, NLP)
DeepLearning.AIDeep Learning Specialization
DeepLearning.AIGANs Specialization
DeepLearning.AITensorFlow Advanced Techniques
Writing & Community
links