Vibhuti Gandhi

Data Scientist · ML Engineer · Applied AI

I build production-grade data science systems — from Bayesian models and LLM pipelines to enterprise CLV engines — that move real dollars and drive real decisions.

📍 Vancouver, BC

🎓 B.Sc. Data Science, Simon Fraser University

🔗 LinkedIn

📧 Email: gandhivibhuti1802@gmail.com

💼 Experience

Data Scientist (Co-op → Full-Time Contract) — British Columbia Lottery Corporation

Jan 2025 – Present

Led end-to-end development and enterprise-wide deployment of advanced analytics solutions at BCLC’s Advanced Analytics division.

Player Segmentation: Developed a behavioral segmentation model using Discrete-Time Markov Chains, deployed to the Player Rewards program — identified $400K/week in incentive inefficiencies and enabled capital reallocation.
Customer Lifetime Value: Co-built a CLV model that improved revenue forecasting accuracy by 20%; operationalized via Snowflake to optimize a multi-million-dollar player journey budget.
Served as a thought leader in the AI & Data Science Community of Practice (COP), publishing articles and presenting to champion ML adoption across the business.

Data Science Research Assistant — SFU MAGPIE Group

Jan 2023 – Dec 2023

Built a multi-source COVID-19 dataset spanning 130+ metrics; applied ML (random forest, gradient boosting, hierarchical clustering) to model Omicron variant evolution.
Co-authored research on COVID-19 variant dynamics; findings were presented at the 31st International Dynamics & Evolution of Human Viruses conference.

Data Analyst Co-op — BC Public Service: Ministry of Health

Jan 2023 – Apr 2023

Built Oracle-to-Power BI data pipelines for real-time physician shift reporting; created dynamic dashboards with drill-down maps and time-series charts.
Resolved data discrepancies in collaboration with the policy team, achieving a 15% reduction in reporting errors.

🚀 Featured Projects

Bayesian Media Mix Modelling for Global Marketing Attribution

PyMC Bayesian Inference SciPy Optimisation Hierarchical Modelling · ▶ Read the Article

Built a full-stack MMM pipeline for a fictional outdoor apparel brand across five markets. Generated synthetic data with known ground truth, applied geometric adstock and Hill saturation transformations, and fit a hierarchical Bayesian model to decompose revenue by channel. A SciPy budget optimizer identified a consistent reallocation from Social → Paid Search yielding an 8–15% weekly revenue uplift across all regions.

Regime-Switching Portfolio Optimisation with Hidden Markov Models

Gaussian HMM Ledoit-Wolf Shrinkage Walk-Forward Validation Canadian ETFs · ▶ Read the Article

Built a three-regime market detection system using a Gaussian HMM on 10 engineered features (realized volatility, momentum, credit spread proxy, yield curve slope), combined with mean-variance optimization. Achieved a Sharpe ratio of 1.27 vs. 0.88 for a 60/40 benchmark, outperforming in 3 of 4 out-of-sample test periods.

Multi-Agent NLP Analysis of TikTok Content

LLMs Gemini-2.5-flash Multi-Agent Architecture NLP

Scraped and analyzed 2,500 videos from 25 creators. Developed a custom Authenticity Index and deployed a multi-agent LLM architecture (Gemini-2.5-flash) to quantify how sponsorship alters linguistic patterns and how audiences respond to those changes.

🏆 Hackathon Winner: Canada’s CPI During the Pandemic

ARIMA Time-Series Python · ▶ Watch the video

Implemented an ARIMA model to forecast Canadian inflation with 92% accuracy during the pandemic — earning 1st place at Vancouver Datajam. Delivered actionable insights into consumer price dynamics during crisis periods.

🧪 Other Projects

Project	Methods	Highlights
Explainable GNNs for Protein Classification	PointNet, GCN, GGS-NN	Integrated explainability analysis for trustworthy geometric deep learning
Diabetes Prediction: Binary Classifiers & SVMs	LR, KNN, LDA, SVM (RBF)	Comprehensive model comparison on Pima Indians dataset
Dow Jones Time-Series Analysis	ARIMA, Box-Jenkins	Addressed volatility clustering, heteroskedasticity, mean reversion
Bank Churn Analysis	Azure Synapse, MySQL, Power BI	Interactive 4-year churn dashboard
Vancouver Weather Forecast	Neural Prophet, PyTorch	78% accuracy using 80 years of climate data
Wikipedia Summarizer GPT App	RAG, Flan-T5 XXL, Streamlit	End-to-end NLP app hosted on Streamlit

🛠 Skills

Languages: Python · R · SQL · C · C++ · MATLAB

ML / AI: PyTorch · TensorFlow · Keras · Scikit-learn · XGBoost · LightGBM · LangChain · Hugging Face · PyMC

Specializations: Bayesian Modelling · Time-Series Forecasting · NLP & Generative AI · Deep Learning (CNNs, RNNs, GNNs) · Customer Segmentation · CLV · A/B Testing · Geospatial Analysis · Reinforcement Learning

Data & Cloud: Azure · AWS · Databricks · PySpark · Snowflake · Oracle · PostgreSQL · MySQL

MLOps: CI/CD · Docker · Azure ML Studio · DataIku · Git/GitHub · Model Monitoring

Visualization: Power BI · Tableau · Plotly · Streamlit · Seaborn

🎓 Education

BSc Data Science — Simon Fraser University (Sep 2020 – Dec 2025)

Relevant Coursework: Object Oriented Programming, Applied Multivariate Analysis, Sampling and Experimental Design, Statistical Learning and Prediction, Bayesian Statistics, Artificial Intelligence, Time Series Analysis, Object Oriented Programming, Database Systems, Linear Optimization

📜 Certifications

✅ Microsoft Certified: Azure AI Fundamentals
✅ DataIku: ML Practitioner
✅ DataIku: MLOps Practitioner
✅ DataIku: Generative AI Practitioner
✅ DataIku: Advanced Designer · Core Designer · Developer
✅ Kaggle: Intro to Deep Learning · Computer Vision · Geospatial Analysis
✅ Udemy: Machine Learning A-Z · PySpark Essentials

🎤 Teaching & Mentorship

Workshop Organizer: House Price Prediction — led 20+ participants through MLR, regularization, and exploratory data analysis with DSSS.
Mentor: Word Embeddings (ELMo & BERT) — coached participants on contextual vector representations applied to climate and disaster event analysis.