Min Htoo Lin — Lead ML Engineer

https://github.com/linminhtoo ML Researcher and Engineer with 4 years of academic and industrial experience. I love combining data, domain knowledge and clever ML algorithms to tackle the most difficult problems with the highest potential to benefit humanity. In particular, I am excited by AI 4 Science. Having done wet-lab research for more than 3 years in university before self-learning ML, I am passionate about creatively applying ML methods to accelerate scientific research and problem-solving. I have accumulated deep experience in rapidly iterating technologies, with a strong focus on ML, but also Software Engineering & Automation, to solve business problems. I also have extensive research and academic experience during my university years, including a 1 year full-time stint at MIT with Prof Connor Coley conducting AI research. These valuable attachments have equipped me with creative problem solving skills and perseverance, as research never goes smoothly and required me to constantly think out of the box. In my free time, I love learning new things. I also enjoy lifting weights at the gym, and climbing stairs for a good cardio workout. Since young, I've been a fan of Korean culture, and I've recently been learning the language to an advanced level. I'm also an avid photographer. My next goal is to learn the piano so that I can play my favorite songs :) I also try to give back as both a mentor to juniors, and as a contributor to community initiatives. I aspire to mentor the next generation of leaders and doers who will lead revolutions of their own.

Stackforce AI infers this person is a Healthcare and SaaS expert specializing in AI-driven scientific research and machine learning applications.

Location: Singapore, Singapore

Experience: 8 yrs

Skills

Applied Machine Learning
Software Engineering
Drug Discovery
Deep Learning
Computer Vision
Scientific Research
Time Series Forecasting

Career Highlights

Achieved 99% accuracy in drug discovery models.
Led a team to integrate quantum physics with ML.
Published award-winning research in cheminformatics.

Work Experience

Stealth AI Startup

Applied AI Engineer (2 yrs 2 mos)

QDX

Lead Machine Learning Engineer (1 yr 11 mos)

ADVANCE.AI

Deep Learning Researcher (Computer Vision & Natural Language Processing) (9 mos)

Massachusetts Institute of Technology

Deep Learning Researcher (Automating Chemistry) (10 mos)

Competitions

Data Science Competitor (3 mos)

CN Yang Scholars Programme

Chairperson, Camp Solve 2019 (10 mos)

Researcher (Organic Chemistry) (2 yrs 4 mos)

Education

Master's at Georgia Institute of Technology

BSc at Nanyang Technological University Singapore

Bachelor of Science - BS at CN Yang Scholars Programme

Bachelor of Science - BS at University of Waterloo

GCE A Levels at Victoria Junior College

GCE O Levels at Victoria School (Singapore)

Min Htoo Lin

Lead ML Engineer

Singapore, Singapore8 yrs experience

Most Likely To Switch

Key Highlights

Achieved 99% accuracy in drug discovery models.
Led a team to integrate quantum physics with ML.
Published award-winning research in cheminformatics.

Stackforce AI infers this person is a Healthcare and SaaS expert specializing in AI-driven scientific research and machine learning applications.

Contact

linmin.htoo@gmail.com LinkedIn

Skills

Core Skills

Applied Machine LearningSoftware EngineeringDrug DiscoveryDeep LearningComputer VisionScientific ResearchTime Series Forecasting

Other Skills

Adversarial LearningCNNsCollaborative Problem SolvingCommunicationComputational ScienceCreative Problem SolvingCreativityCritical ThinkingData AnalysisData ScienceData augmentationsDataset curationEfficientNetEnergy-Based ModelsEquivariant Graph Neural Networks

About

Experience

8 yrs

Total Experience

1 yr 7 mos

Average Tenure

2 yrs 2 mos

Current Experience

Stealth ai startup

Applied AI Engineer

Apr 2024 – Present · 2 yrs 2 mos · Singapore, Singapore · On-site

(company name hidden for confidentiality)
Architected interactive platform for complex, domain-specific data processing and ML workflows. Developed in Backend (Python, Pytorch, Scikit-Learn) and Frontend (NiceGUI, Javascript). Built for rapid iteration and reproducibility, our platform enabled data scientists to deliver projects to clients in days instead of weeks.
Researched ML techniques to address data issues using Graph Neural Nets (GNNs), Variational Autoencoders (VAEs), Adversarial Learning and Self-Supervised Learning, improving cross-validation metrics by as much as 2x.
Built production-grade libraries for domain-specific machine learning: data processing, training, inference, visualization & delivery. Enforced good coding practices, CI/CD & code review.
Optimized workflows such as model training from 24 hours to 20 minutes and model inference from 20 hours to 30 minutes, by profiling bottlenecks and employing strategies such as TensorRT compilation and parallelization.
Developed tool to rank features with SHAPley values, which accelerated workflows by 50x while preserving accuracy.

PyTorchCreative Problem SolvingResearchSoftware EngineeringSystems DesignApplied Machine Learning

Qdx

Lead Machine Learning Engineer

Apr 2022 – Mar 2024 · 1 yr 11 mos · Singapore, Singapore · Hybrid

Solely spearheaded research into Equivariant Graph Neural Networks to map 3D molecular structure to complex quantum chemical properties. Performed dataset curation, model design, training and validation. Achieved 99% agreement with high fidelity simulations while being >1000x faster, saving 120K+ $/yr in compute costs. Executed 100k+ simulations on 500+ GPUs to curate one of the world’s largest quantum property datasets at that time.
Developed generative algorithm to design potential drugs using Markov Decision Processes & Genetic Algorithms while satisfying multiple constraints.
Led 3 engineers in integrating quantum physics simulation with machine learning to identify drug candidates better than competitors by 200% in Enrichment Factor. This was crucial for securing our seed funding as a startup.
Read research on geometric machine learning and developed in-house models to predict drug-protein interactions.

LeadershipDrug DiscoveryGraph Neural NetworksHigh Performance Computing (HPC)PyTorchResearch and Development (R&D)+1

Advance.ai

Deep Learning Researcher (Computer Vision & Natural Language Processing)

Jun 2021 – Mar 2022 · 9 mos · Singapore, Singapore · Hybrid

Researched Transformer model for document layout understanding with a novel training data generation pipeline. Required <5 human-labeled examples to classify text entities with >98% accuracy. Collaborated with PMs & engineers to deploy cutting onboarding time of our global, multi-lingual enterprise clients from 3 weeks to 1 day
Trained YOLO-based forgery detection models, improving recall from 60% to 97% at 1% false rejection by analyzing failure cases & crafting data augmentations to address them.
Developed a Super-Resolution method to enhance text recognition accuracy by 10 percentage points using a text recognition alignment loss. By training on synthetically augmented data, the method required no human labeling.

ResearchProblem SolvingTeamworkComputer VisionDeep LearningPyTorch+2

Massachusetts institute of technology

Deep Learning Researcher (Automating Chemistry)

Aug 2020 – Jun 2021 · 10 mos

Developed practical ML methods for the pharmaceutical industry. Reframed prediction of chemical synthesis pathways as two-stage problem of generation + ranking. Trained Energy-Based Models via Semi-Supervised Learning to rank candidate pathways. Improved Recall@1 & Mean Reciprocal Rank by 45% over academic baselines.
Published 1st-author paper in the Journal of Cheminformatics (29 citations), contributing most of the coding, model experimentation, data analysis & writing. Open-sourced the code and data.
Won national research excellence prize, with Prof. Connor Coley’s nomination (see file)-

Data ScienceProblem SolvingDeep LearningPyTorchWritingScientific Research

Competitions

Data Science Competitor

May 2020 – Aug 2020 · 3 mos · Singapore · Remote

Energy Demand Forecasting Competition, ai4impact
Built neural network with 3 teammates to predict a building’s energy demand 1 day ahead, beating the baseline Root Mean Squared Error by 53% & winning 1st out of 50 teams. Published Editor’s Pick article in TowardsDataScience, with 25k+ reads, explaining our literature research & time-series feature engineering: https://www.tinyurl.com/mediumcircuit
Wind Energy Trading Competition, SGInnovate
Developed time-series deep learning models fusing CNNs & LSTMs to trade wind energy with 3 teammates.
Emerged 1st of 100 teams in profits after 7 days of live trading simulation. Article: https://www.tinyurl.com/stonksmedium
Shopee Code League x Kaggle: Image Classification Competition
Built EfficientNet model with 1 teammate to classify 110k e-commerce product images into 42 classes. Placed top 3% among 823 teams in Asia by implementing SOTA techniques from literature to tackle noisy labels and overfitting.

CreativityResearchDeep LearningRapid PrototypingCollaborative Problem SolvingPyTorch+1

Cn yang scholars programme

2 roles

Chairperson, Camp Solve 2019

Promoted

Aug 2018 – Jun 2019 · 10 mos

• Led 25 members, managed $6,000 budget & negotiated with 8 NTU faculties to initiate a 3-day camp to expose 40 poly students to STEM research; gained NTU management’s buy-in to continue the camp for subsequent years

Researcher (Organic Chemistry)

Jan 2018 – May 2020 · 2 yrs 4 mos

Co-authored publication (top downloaded paper of 2019) in Asian Journal of Organic Chemistry, 2019, 8(7), 1058-1060
Collaborated with PhD students & post-docs to develop cheaper, greener & non-toxic versions of traditional reactions. Contributed to 5 research projects across 2 research groups
Compiled an extensive 100-page literature report now used as a reference by the group
Awarded Best Research among 50 CN Yang scholars, presented in SG Chem National Meeting & ACS Undergrad Symposium