Best fit
Data scientist or machine learning engineer roles focused on practical AI systems.
Vancouver, BC, Canada / Data Scientist | Machine Learning Engineer
Mandeep
I build practical AI systems for operations, recommendations, scientific data, and language-heavy workflows. My work spans mathematical optimization, time series forecasting, LLM/RAG products, privacy-preserving synthetic data, and applied research.
Current Focus
Practical AI systems that connect models to decisions.
Strongest domains
Optimization, forecasting, recommender systems, LLM/RAG products, and applied research.
Working style
Translate ambiguous operational problems into measurable models, pipelines, and decision support.
Optimization work translating supply chain rules, capacity limits, routing choices, and demand priorities into solver-ready formulations.
Active learning pipeline for satellite imagery selection with SpaceML and NASA IMPACT.
Research initiative ranking for AI-assisted Earth observation workflows.
Explainable AI, satellite data curation, mobile ML for agriculture, and machine-learning forecasting.
Selected Work
Operational ML with measurable constraints.
Data Scientist Canfor
Optimization, forecasting, and enterprise ML workflows for forestry supply chain and manufacturing operations.
- Leading a large-scale logistics optimization program that turns operational constraints into mathematical formulations for lower shipping and routing costs.
- Built solver-agnostic optimization pipelines across SciPy HiGHS, SCIP, CPLEX, and Gurobi to compare solution quality, runtime, and scalability.
- Developing production forecasting pipelines with Chronos2, TSMixer, XGBoost, LightGBM, and unit8-darts, using rigorous backtesting and error analysis.
- Deploying and orchestrating ML workflows with Azure Machine Learning and Microsoft Fabric for reproducible enterprise experimentation.
Machine Learning Engineer CoffeeSpace
Recommendation systems, feature engineering, analytics pipelines, and cloud-native deployment.
- Enhanced a people-to-people recommender system with LLM-driven feature engineering and retrieval-augmented generation.
- Built analytics pipelines to evaluate user behavior and recommendation quality with BigQuery and Looker Studio.
- Developed Python, JavaScript, Firebase, and GCP Cloud Run services to support recommendation and analytics workflows.
- Integrated Qdrant vector search to improve similarity retrieval and personalization.
Machine Learning Engineer Betterdata
Privacy-preserving synthetic tabular data generation and evaluation.
- Researched and implemented generative model architectures for synthetic data diversity, fidelity, and privacy.
- Integrated differential privacy into PyTorch-based generative models to support stronger data protection standards.
- Optimized training and evaluation pipelines with Numba for cloud and on-premises deployment.
- Engineered evaluation systems for relevance, statistical fidelity, accuracy, and privacy of generated data.
AI Researcher SpaceML
Self-supervised and active learning for petabyte-scale satellite imagery.
- Collaborated with NASA IMPACT to identify relevant Earth observation imagery across massive unlabeled datasets.
- Built scalable active learning pipelines that reduced manual image labeling from 7,000 hours to 52 minutes for a five-million-image climate dataset.
- Open-sourced labeling tooling used in NASA's Phenomenon portal.
- Contributed to a NASA Science Mission Directorate grant proposal ranked in the top 5 of 79 initiatives.
Research Assistant Thapar University
Explainable AI, medical image synthesis, and applied forecasting research.
- Studied CNN architecture choices and hyperparameter effects on explainability methods including LIME and SHAP.
- Explored GAN-based synthetic COVID-19 CT and MRI generation to improve deep learning detection systems.
- Built machine-learning forecasting workflows for social media traction using NumPy, Pandas, SQL, and model selection.
Projects
Optimization models, research prototypes, and deployed AI workflows.
Applied mathematics Current
Supply Chain Optimization Engine
A large-scale linear programming and operations research project for modeling shipping, capacity, routing, and demand-priority decisions in forestry operations.
- Context
- Forestry supply chain planning with capacity limits, demand priorities, and routing tradeoffs.
- Role
- Own mathematical modeling, solver comparison, and translation of operational rules into solver-ready formulations.
- Outcome
- Projected annual logistics savings of approximately $1.5M.
Product ML Product
People-to-People Recommendation System
A matching and personalization workflow using LLM-derived features, RAG context, vector search, and behavioral analytics.
- Context
- Cloud-native recommendation and analytics workflows for a people-matching product.
- Role
- Enhanced feature engineering, retrieval, analytics, and deployment workflows.
- Outcome
- Improved recommendation evaluation and similarity search for product workflows.
NLP research Research
Biomedical Lay Summarization with LLMs
Generated accessible biomedical research summaries for non-expert audiences using domain-specific LLMs, prompt tuning, RAG, and representation engineering.
- Context
- Biomedical NLP workflow for making technical research easier for non-expert readers.
- Role
- Designed summarization prompts, retrieval context, and representation-control experiments.
- Outcome
- Improved factuality, readability, and accessibility of complex biomedical summaries.
LLM reasoning Research
Annotated Corpus for Reasoning Explanations
Created structured natural-language explanations for AI2 Reasoning Challenge questions and used Tree of Thought prompting to guide step-by-step problem solving.
- Context
- Science-question reasoning workflow for improving transparency in model outputs.
- Role
- Built explanation annotations and experimented with structured reasoning prompts.
- Outcome
- Improved transparency and reasoning quality for complex science-question workflows.
Earth observation Open source
SpaceML Worldview Search
Active learning system for identifying relevant satellite imagery from petabyte-scale unlabeled Earth observation datasets.
- Context
- Citizen-science and NASA IMPACT collaboration for large-scale Earth observation data curation.
- Role
- Built active learning and labeling workflows for surfacing useful imagery from massive unlabeled datasets.
- Outcome
- Reduced labeling time from 7,000 hours to 52 minutes for a five-million-image climate dataset.
Mobile ML Published
Grape Leaf Disease Diagnosis
Real-time iOS app for grape leaf disease diagnosis with on-device deep learning, remedy suggestions, and model optimization.
- Context
- Mobile agricultural diagnosis system designed for real-time field use.
- Role
- Led mobile ML implementation, model optimization, and on-device inference integration.
- Outcome
- Optimized models with quantization and pruning for iPhone 8 and newer devices.
Publications
Peer-reviewed work across explainable AI, Earth science, mobile ML, and forecasting.
Deep learning-based explainable target classification for synthetic aperture radar images
Explainable deep learning for SAR target classification, combining model performance with interpretable evidence for high-stakes imagery workflows.
SpaceML Worldview Search - Learnings from an AI citizen scientist team building a NoCode Data Curator from Unlabeled Petabyte Scale Imagery
Lessons from citizen-science tooling for surfacing useful imagery from petabyte-scale unlabeled Earth observation data.
Smartphone Based Grape Leaf Disease Diagnosis and Remedial System Assisted with Explanations
Mobile explainable AI for grape leaf disease diagnosis, built for real-time on-device use and practical field support.
Machine Learning Based Explainable Financial Forecasting
Explainable forecasting research focused on making model signals more transparent for financial prediction workflows.
Paper
First author
Capabilities
A practical toolkit for turning models into working systems.
Optimization and forecasting
- Linear programming and mixed-integer optimization
- SciPy HiGHS, SCIP, CPLEX, and Gurobi
- Chronos2, TSMixer, XGBoost, LightGBM, and unit8-darts
- Backtesting, model comparison, and error analysis
Machine learning systems
- PyTorch, TensorFlow, Keras, Hugging Face, and Scikit-learn
- Recommendation systems and vector search
- Synthetic data generation and differential privacy
- Active learning and self-supervised learning
Language and data products
- LLMs, RAG, prompt engineering, and control vectors
- Natural language processing and computational linguistics
- BigQuery, PostgreSQL, MongoDB, Firebase, and SQL
- Looker Studio, Plotly, Altair, Matplotlib, and Seaborn
Engineering and delivery
- Python, R, JavaScript, Bash, and MATLAB
- Docker, GitHub Actions, Git, GitHub, and GitLab
- Azure ML, Microsoft Fabric, GCP Cloud Run, AWS, and Kubernetes
- Weights & Biases, experiment tracking, and reproducible workflows
Education
Data science, computational linguistics, and computer engineering.
Master of Data Science in Computational Linguistics
University of British Columbia / Vancouver, BC
GPA 93.4. Focused on advanced NLP, transformer models, machine learning and optimization, computational linguistics, and interactive data visualization.
Bachelor of Technology in Computer Engineering
Thapar Institute of Engineering and Technology / Patiala, Punjab, India
GPA 84.5. Covered data structures and algorithms, AI and machine learning, database systems, and software engineering.
Contact