Home

Software engineer · AI, ML & 3D algorithms

Daniel ChenKwok Lee

I build the parts of AI systems that have to survive contact with real users — agents that book real appointments, retrieval that returns the right context, and 3D algorithms running inside surgical planning software. MSc Computer Science, University of Nottingham.

Email me Visualize my Research
7 yearsShipping software · Dev & QA
MSc, MeritNottingham · AI & ML
Agentic AIIn production, not demos
Python · C++ · C#Plus SQL, MDX, R

Work

Seven years of making things work in production.

Medical 3D imaging today, agentic AI and retrieval before that, and a long stretch of machine learning and data engineering before that. Newest first.

2025 — present

Software Development Engineer

Materialise Sdn. Bhd. · Selangor

Malaysian arm of Materialise NV (Belgium) — high-precision 3D printing and mass customisation for manufacturing and medical clients.

  • Write 3D image and mesh processing algorithms in a proprietary Python framework, so clinical cases are prepared automatically instead of by hand.
  • Build the Python plugin GUI clients use alongside Mimics and 3-matic, letting them scale case volume without scaling headcount.
  • Turn customer reviews into requirements, then into the algorithm that satisfies them.
  • Started an internal initiative for an AI agent that reviews code and tracks code quality, powered by Materialise's OpenAI model.
Python3D geometryMedical imagingGUILLM agents
2024 May — Nov

Software Engineer

Athena AI (Pand.Ai Pte. Ltd) · Kuala Lumpur

Malaysian subsidiary of a Singaporean AI startup — on-premise LLM deployments, LLM customisation and NLP chatbots.

  • Shipped the agentic layer of a major client's pilot: an OpenAI-powered agent that retrieves a customer's past transactions and books their appointments.
  • Improved answer quality with RAG over a Weaviate vector database, then personalised replies per user from transaction profiles in MongoDB.
  • Kept Dockerised chatbots healthy on Kubernetes across several clients — both LLM and classical NLP stacks.
  • Mentored a new teammate through onboarding, from troubleshooting to system familiarity.
Agentic AIRAGWeaviateMongoDBDockerKubernetes
2020 — 2023

Software Engineer

Nettium Sdn. Bhd. & DataBricks Tech Sdn. Bhd. · Kuala Lumpur

Online gaming, e-commerce, gamification and talent management software. The BI department was acquired and spun out as DataBricks Tech; I moved with it.

  • Built the company's first internal AI specialist — a prototype embedding model plus custom context, prompts and logic — used for staff training and troubleshooting.
  • Reworked a Monte Carlo Markov Chain model for customer behaviour prediction; proprietary modifications cut prediction variance and raised accuracy.
  • Owned the backend Python behind customer profiling: pipelines pulling from SQL Server, PostgreSQL, MDX/OLAP cubes and APIs, with error tolerance built in.
  • Automated the team's tests and alerting into Teams and email, and ported R and C++ libraries into Python.
  • Supported a widely-used C# API that other teams relied on for MicroCube, OLAP and ROLAP data.
PythonMCMCSQL ServerPostgreSQLMDX / OLAPC#Prompt engineering
2018 Aug — Sep

Pro bono AI consultant

Jeff Leong, Poon & Wong · Kuala Lumpur

A Legal 500 Asia Pacific firm working on corporate governance and M&A.

  • Interviewed the firm's lawyers to find which parts of legal work were genuinely automatable — and which were not.
  • Vetted vendors and recommended a cloud provider for legal automation, free of charge.
AI strategyStakeholder researchVendor selection
2015 — 2018

Business & Integration Architecture Analyst

Accenture Solutions Sdn. Bhd. · Kuala Lumpur

Fortune Global 500 consulting, technology and outsourcing.

  • Ran quality assurance for Accenture NewsPage — a distributor management platform with 160,000+ users, 5,000 distributors and 3 million retail outlets across 60 countries.
  • Organised testers across Scrum sprints, coordinated stand-ups with developers, and joined requirements planning to design logical errors out before they were built.
  • Held regression testing across up to five live versions at once, using SQL and Git daily and fixing JavaScript failures in Selenium.
  • Semi-automated daily ticket reporting for a major telco, saving roughly four hours of manual work a week, and traced the biggest drivers of customer dissatisfaction back to root causes for a broadband provider.
QA strategyScrumSeleniumSQLRoot-cause analysis

Research

Teaching a search how to get unstuck.

Adding Vector Sampling to the Reinforcement Learning Artificial Bee Colony (R-ABC) Algorithm for Feature Selection — MSc thesis, University of Nottingham

Optimisers get trapped. They climb the nearest hill, run out of better moves, and mistake the top of that hill for the top of the map. That is a local optimum, and it is the reason a model can look converged while being quietly wrong.

My thesis adds a new escape to the R-ABC metaheuristic. When a scout has failed too many times, it does not restart at random. It takes the midpoint of everywhere the colony has already searched and launches a random Euclidean vector of a chosen length from it. The longer the vector, the lower the chance of landing on ground already covered — so search distance becomes a dial you can turn.

It works as a convergence and feature selection strategy for neural networks, decision trees and SVMs. Watch the panel: the scout gets stuck, gives up, jumps, and finds the real peak.

A scout starts on unsearched ground.

Toolkit

Strongest

Daily drivers. I design in these.

PythonAgentic AI systemsC++C Machine learningRAG & embeddingsPrompt engineering

Working knowledge

Shipped real features with these.

RJavaC#MATLAB GitMongoDBPython GUIDockerKubernetesWeaviate

Familiar

Enough to be useful and to get fluent fast.

SQLMDX / OLAPJavaScriptSelenium

Sharpening

What I'm deliberately getting better at.

Decision-making algorithmsNew languages, on purpose

Education

MSc Computer Science

University of Nottingham · 2018–2019

Merit (≈3.7/4.0). Artificial Intelligence, Machine Learning, and a research project in Computer Science.

BEng (Hons) Electronic & Computer Engineering

University of Nottingham · 2012–2014

Second Class, First Division (≈3.7/4.0). Thesis: SVM on Arduino — supervised learning doing real-time sensor classification on constrained hardware.

GCE A-Levels · 4A*

Methodist College Kuala Lumpur · 2010–2011

Mathematics, Physics, Chemistry, Biology.

Bishop Denis C. Dutton Scholar 2011 — best A-Level performance

SPM · 12A (10A+)

Sekolah Menengah Sri Sempurna · 2005–2009

Also passed ACCA papers F1–F5 and F9, so finance stakeholders and I speak the same language. English and Malay, both fluent.

Top SPM student, 2009

Contact

Let's build something that ships.

Based in Kuala Lumpur, comfortable with distributed teams, and interested in problems that are harder than they first look. If you have work that needs someone who ships, say hello — I reply to every message.

Emailstudyhard3@yahoo.com