Hey! My name is Andre. I'm a final year studying Mathematics & Computer Science at the National University of Singapore (NUS).
Machine learning is an exciting field at the intersection of mathematical theory (brutal math courses finally paying off.. *_*) and software engineering. There's no shortage of groundbreaking research in the ML landscape, and I aspire to be among those who bring these innovations into real-world applications.
I am now particularly interested in the intricacies of parallelism in training/inference optimization and distributed systems. I aspire to design the infrastructure of next-generation ML systems and pipelines.
Beyond academia, I am a casual climber (an occasional diver, and avid backpacker) and I am part of the university's Mountaineering club and Climbing club. Together with a couple of πΈπ°π―π₯π¦π³π§πΆπππΊ π§πΆπ―-ππ°π·πͺπ―π¨ π€πΆπ€π¬π°π°π΄, we scaled the Himalayas and it was simply fantastic!
Delivered as my Final Year Thesis project. A multi-agent system that integrates directly into Overleaf to help authors with LaTeX-aware debuging and revising academic papers. I oversee our custom (currently close-sourced) backend orchestration logic β XtraMCP. It is both a research contribution and an engineering system that I, alongside few others, are now advancing toward a production-grade platform.
Part of Prof He Bingsheng's research group, focusing on distributed ML systems and adaptations of the transformer architecture. Fortunate enough to make some publications along the way!
My stint as a CS2040S (Discrete Structures & Algorithms) has convinced several capable and passionate ex-students of mine to join me in developing an open-source teaching material for future cohorts. Lovely!
The bank was in its Agentic AI phase, so I learnt, built, and extended custom MCP integration and validation logic for backend services.
Software Engineering meets Quantitative Trading - Learnt how to support the trading team. Taught me zero-tolerance engineering.
Gained practical knowledge on system design and was taught what simple, reliable, sustainable, and fault-tolerant systems look like.
Worked on finetuning LLMs using data and model parallelism techniques to contend with larger models at lower cost. Also learnt to design, build, and deploy ML pipelines in production. I trace my origins here.
Learnt ML production and deployment lifecycle, and worked on Quant Research projects affiliated with QRT.
Teaching Assistant for CS1010s (Programming Methodology in Python) and CS2040s (Data Structures and Algorithms); Won a teaching excellence award!
Bustling city. Classy heels. Dazzling lights. Furious pace. Utter shit.
The seeds are beginning to sprout
Toska: An Unexpected Yearning
Attention Mechanism behind LLMs - One of the more enlightening moments from this hectic semester
Python, Java, C++, TypeScript
TensorFlow, PyTorch, Scikit-learn, OpenCV, pandas, NumPy
vLLM, llama.cpp, DeepSpeed, Lightning AI, LitGPT, LangChain
PostgreSQL, Spark, Flink, Kafka
Docker, Django, FastAPI, Spring Boot, Express, Nodejs