Voon Yan Kho
Open to roles

Voon Yan Kho,
data scientist.

Data Science & Computer Science graduate at the University of Western Australia. I work at the intersection of data engineering, machine learning, medical research, and social impact. Based in Perth, WA.

currently focused on  → dbt & Snowflake pipelines diffusion MRI tractography deep learning in PyTorch dbt & Snowflake pipelines
Kaggle ↗ LinkedIn ↗ Resume ↗ Email →
About

I build with data - from clinical MRI scans to charity registries to health-insurance analytics. I like problems where the answer changes how a person, a policy, or a pipeline actually behaves.

I'm a Data Science and Computer Science graduate at UWA. I've held research and data engineering roles at HBF, KEMH's Neonatal Health team, the UWA Centre for Social Impact, and the Harry Perkins Institute of Medical Research: spanning dbt & Snowflake, diffusion MRI, web scraping, and single-cell RNA sequencing.

I hold 17 IBM certifications: a certified IBM Business Intelligence Analyst, and currently completing the IBM AI Engineering Professional Certificate. I speak English, Chinese, and Malay. Outside work: volunteering, travel, roadtrips, and good coffee.

Projects all projects ↗
🌦️
Australia Weather Data Pipeline — End-to-End ELT
A self-refreshing data engineering project collecting daily weather for Australia's eight capital cities from the Open-Meteo API. Incremental ingestion (year-one backfill, then a rolling 10-day refetch) lands raw JSON, an ETL step upserts a tidy CSV source-of-truth, and a DuckDB warehouse is rebuilt with a fact table, city dimension, and monthly / city-summary mart views. A GitHub Actions cron runs it daily and commits fresh data, feeding an interactive Streamlit dashboard.
Data Engineering · DuckDB · ETL · GitHub Actions
↗
🌡️
Australian Capital Cities Weather — EDA & Prediction
Built on data from my own weather pipeline. Loads the live CSV straight from GitHub, runs leakage-safe feature engineering (calendar, lag & rolling features), a deep EDA of temperature, rain, wind & sunshine, then predicts tomorrow's max temperature and rain with classical models and an LSTM — all judged honestly against persistence baselines on a strictly chronological split. Boosted trees clearly beat the baselines on rain; temperature stayed hard to beat, and I say so plainly.
Data Science · LSTM · Forecasting · Time Series
↗
✈️
Global Aviation Industry Analysis & Prediction
A full data-science cycle across five aviation datasets (2010–2026): airline financials, Boeing vs Airbus order books, regional passenger traffic, top routes and major incidents — covering the 737 MAX crisis, COVID collapse and recovery. Cleaning, feature engineering, EDA, nine analytical tasks, then machine learning and an LSTM forecasting monthly regional traffic, with the LSTM topping the leaderboard by learning momentum through the out-of-range recovery period.
Data Science · EDA · LSTM · Forecasting
↗
🎣
Phishing Email Detection with Deep Learning
Classifies 82,000+ emails from six merged public corpora as phishing/spam vs legitimate. Benchmarks Naive Bayes & Logistic Regression on TF-IDF against three deep models — a feedforward net, a 1D CNN, and a bidirectional LSTM — selecting on spam-class F1. Key finding: on pre-cleaned text, word identity beats word order, so TF-IDF models won; an interpretability audit even exposed corpus-identity leakage that caps how far the benchmark should be trusted.
Deep Learning · NLP · BiLSTM · Text Classification
↗
⚙️
Crypto Market Data Pipeline — End-to-End ELT
A self-updating data engineering project: pulls hourly crypto prices from the CoinGecko API with incremental, watermark-based ingestion, lands immutable raw parquet, then models it into a warehouse with dbt on DuckDB — staging, hourly-to-daily OHLCV, and marts for returns, rolling volatility, drawdown & pairwise correlations. 18 dbt tests + pytest gate every layer, and a GitHub Actions cron rebuilds and commits fresh data daily, feeding a live Streamlit dashboard.
Data Engineering · dbt · DuckDB · GitHub Actions
↗
📉
Deep Learning for Crypto Time Series Forecasting
Built on data from my own crypto pipeline. Trains a global LSTM across 10 coins in PyTorch to forecast next-day log return and next-day volatility, benchmarked against naive, EWMA and tuned ridge baselines. Result reported honestly: returns behave like a random walk (chance-level directional accuracy), while volatility is forecastable — the LSTM posts the lowest MAE/RMSE once anchored on the persistence forecast.
Deep Learning · PyTorch · LSTM · Time Series
↗
🏃
Human Action Recognition — ResNet-50 Transfer Learning
Fine-tuned a pretrained ResNet-50 in PyTorch to classify 12,600 unconstrained web photos across 15 action classes. Stratified 85/15 split, heavy augmentation (crop, flip, jitter, rotation, random erasing), and discriminative learning rates — a gentle 1e-4 on the backbone, 3e-4 on the new head — reaching 84.4% validation accuracy vs 6.7% chance.
Deep Learning · PyTorch · ResNet-50 · Transfer Learning
↗
🖼️
Image Captioning on Flickr8k — CNN + Attention
End-to-end caption generation in TensorFlow / Keras: a frozen InceptionV3 encoder paired with an LSTM decoder guided by Bahdanau attention, trained with teacher forcing and early stopping over the 8,091-image Flickr8k corpus. Evaluated with masked cross-entropy, corpus-level BLEU against 5 human references, beam search, and attention-map visualisation — reaching BLEU-1 = 0.533 / BLEU-4 = 0.139.
Deep Learning · TensorFlow · Attention · NLP + Vision
↗
🚢
Titanic — Machine Learning from Disaster
End-to-end pipeline on the Kaggle classic: imputed Embarked / Age / Cabin, handled outliers, engineered features (log-Fare, Sex × Pclass, Age × Pclass, FamilySize, IsAlone), and benchmarked Logistic Regression, Decision Tree, Random Forest & XGBoost. XGBoost won on ROC-AUC.
Classification · XGBoost · Feature Engineering
↗
🎗️
Breast Cancer Classification — Deep Neural Network
PyTorch feed-forward network (30 → 64 → 2) on the Breast Cancer Wisconsin Diagnostic dataset. Balanced the classes to 200 / 200, standardised 30 features, and trained with Adam + cross-entropy over 10 epochs — final test loss ≈ 0.09. Ran ablations on SGD vs Adam, hidden-unit counts, and generalised the pipeline to the Iris dataset.
Deep Learning · PyTorch · Medical Classification
↗
🍷
Red Wine Quality Prediction
Binary classification on the UCI Vinho Verde dataset (quality ≥ 7 = good). Cleaned 240 duplicates, log-transformed skewed physicochemical features, and benchmarked Logistic Regression, SVM, Decision Tree, Random Forest, and XGBoost across full / balanced / minimal feature sets with class-balanced weights. Random Forest on the minimal set won — leaner features, same accuracy, less overfit.
Classification · XGBoost · Model Comparison
↗
🧠
Red Wine Quality — Deep Learning Extension
Follow-up to the classical ML study, now with TensorFlow / Keras. Built a Dense 64 → Dropout → 32 → Dropout → Softmax MLP with Adam + EarlyStopping (patience 12) and class-balanced weights. Trained on both the full 6-class quality target and a regrouped 3-class (low / medium / high) target across full / balanced / minimal feature sets — the 3-class full model reached ≈ 0.66 accuracy, and misclassifications clustered around adjacent quality levels, confirming the target's ordinal structure.
Deep Learning · TensorFlow · Multiclass
↗
🏘️
Apartment Prices Prediction — Poland
Trained and compared multiple regression models to predict apartment prices across Polish cities. Full EDA, feature engineering, and model evaluation — my first end-to-end ML project.
Regression · scikit-learn · EDA
↗
📈
Income by Education — Australia & Canada
Cross-country comparative analysis of education ROI. Australia offers a higher return on education investment than Canada, especially at higher qualification levels.
Data Analytics · Pandas · Seaborn
↗
🎓
Student Performance Prediction
Predicted students' final grades from prior performance and behavioural factors. Linear regression quietly outperformed more complex models — a useful reminder.
Regression · Feature Engineering
↗
♻️
Waste Classification — IBM Deep Learning Capstone
Automated waste classifier separating recyclable from organic waste using transfer learning and fine-tuning in Keras & TensorFlow. Final project for the IBM AI Capstone.
Deep Learning · Keras · Transfer Learning
↗
🎨
Anime Image Classification — CNN
Built and trained a convolutional neural network from scratch in PyTorch for multi-class anime image classification. Practice project from the IBM PyTorch course.
Deep Learning · PyTorch · CNN
↗
Education
The University of Western Australia
Bachelor of Science — Data Science & Computer Science (double major)
Jul 2024 – Jun 2026
81.4
WAM
6.67 / 7
GPA
17
IBM Certs
3
Languages
Experience four roles · 2024 → 2026
Mar 2026 — Apr 2026
Perth, WA
🏥Data Engineering Intern
HBF Health · health insurance
Built and maintained analytical datasets with dbt and Snowflake. Diagnosed and fixed failing dbt test cases, improving pipeline reliability for the data engineering team.
dbtSnowflakeSQLData Modelling
Nov 2025 — Feb 2026
Subiaco, WA
🧠Research Assistant — Data Science
KEMH Neonatal Health Team · women & newborn hospital
Diffusion MRI tractography of the preterm infant brain in Python & DIPY. Ran the full preprocessing stack — denoising, Gibbs ringing & eddy correction, bias-field correction, brain masking, response-function estimation, and CSD for fiber orientation.
DIPYMRIPythonNeuroscience
Feb 2025 — May 2025
Crawley, WA
🌱Research Intern — Data Science
UWA Centre for Social Impact (CSI) · research institute
Cleaned survey datasets, built Power BI dashboards for social-impact metrics, implemented Whisper / WhisperX for speech-to-text, and scraped a dataset of 65,000+ Australian charities and not-for-profits.
Power BIWhisperWeb ScrapingEDA
Nov 2024 — Jan 2025
Remote
🧬Research Assistant — Data Science
Harry Perkins Institute of Medical Research · biomedical
Single-cell RNA sequencing (scRNA-seq) analysis in Seurat / R. Performed Hallmark GSEA and GO pathway analysis to study differential expression in endothelial cells, and produced ggplot2 reports for collaborators.
RSeuratscRNA-seqggplot2
Skills
🤖Machine Learning
scikit-learn · Bayesian statistics · regression · classification · cross-validation · hyperparameter tuning
🧠Deep Learning
PyTorch · Keras · TensorFlow · CNNs · transfer learning · fine-tuning · neural networks
📊Data Analytics & EDA
Pandas · NumPy · Seaborn · Matplotlib · ggplot2 · data cleaning · feature engineering
⚙️Data Engineering
dbt · Snowflake · ETL · PySpark · Databricks · IBM Cloud · Apache HBase · Hive
🗄️Databases
SQL · PostgreSQL · MySQL · query optimisation · data modelling
📈Business Intelligence
Tableau · Power BI · IBM Cognos Analytics · dashboard design · data storytelling
Certifications IBM · Coursera
Generative AI & LLMs: Architecture & Data PrepFeb 2026
AI Capstone Project with Deep LearningJul 2025
Deep Learning with PyTorchJul 2025
Introduction to Neural Networks & PyTorchJun 2025
Deep Learning with Keras & TensorFlowDec 2024
Introduction to Deep Learning & Neural Networks (Keras)Nov 2024
Practera Study Australia Industry ExperienceOct 2024
IBM Business Intelligence (BI) Analyst SpecializationJul 2024
Oracle Java FoundationJul 2023
Salesforce Lightning Reports & Dashboards SpecialistJun 2023
Contact

Open to graduate roles, research collaborations, and conversations about data. Say hi.

Email wenyan1029@gmail.com ↗ LinkedIn in/voonyan ↗ Kaggle /voonyan ↗ Resume PDF Google Drive ↗
Or send a message