Question answering over 77k chunks of Canadian aviation safety reports in English, French, and Chinese. Combines BGE-M3 dense retrieval in Qdrant with cross-encoder reranking, Neo4j graph entity expansion, and a LangGraph human-in-the-loop review agent backed by FastAPI and Gradio.
Background
I'm an applied machine learning engineer based in Toronto, Canada. Previously at Kinaxis, I built enterprise copilot and multi-agent workflows with LangGraph on Azure and engineered high-throughput data pipelines on Databricks, migrating core batch pipelines from Pandas to Polars for ~75% runtime reductions.
I specialize in end-to-end architectures that bridge modeling and infrastructure — whether that is parsing 77,000 aviation safety document chunks through dense vector retrieval (BGE-M3 in Qdrant) and Neo4j knowledge graphs, or wiring LightGBM forecasting directly into ERP reordering engines.
I hold a B.Eng. in Electrical Engineering from Carleton University and postgraduate credentials in Applied A.I. Solution Development from George Brown College, backed by Azure AI Engineer Associate and AWS Solutions Architect – Associate certifications.
Skills
ML & Agents
- PyTorch
- LangGraph (Agentic Workflows)
- LightGBM & XGBoost
- Hugging Face (TRL, Transformers)
- BGE-M3 & Cross-Encoders
- GRPO / Post-Training
Data & Retrieval
- Polars & Pandas
- Databricks & PySpark
- Qdrant (Dense Vector DB)
- Neo4j (Knowledge Graphs)
- PostgreSQL & SQLite
- Kafka & Redis
Cloud & Systems
- Microsoft Azure (Certified)
- AWS (Certified Solutions Architect)
- FastAPI
- Docker & Kubernetes
- CI/CD & GitHub Actions
- Odoo 17 ORM & OWL JS
Featured
Production Odoo 17 module that trains LightGBM regressors on historical sales order series and writes 7-day demand forecasts directly into native inventory reordering rules (`stock.warehouse.orderpoint`). Includes an interactive OWL dashboard embedded in Odoo Inventory.
Contributed engineering fixes to the vision-language GRPO (Group Relative Policy Optimization) recipe for modern TRL: conversational chat templating, reward function alignment, and pinned stable dependencies.
A Guitar Hero-style rhythm game for physical Novation Launchpad controllers, converting real-time MIDI input streams into synchronized audio-visual gameplay.
Experience
Built enterprise copilot and multi-agent workflows using LangGraph and Azure OpenAI, incorporating human-in-the-loop review checkpoints. Engineered large-scale data pipelines on Databricks, migrating core batch workloads from Pandas to Polars to cut execution runtimes by ~75%.
Archive
| Year | Title | Category | Links |
|---|---|---|---|
| 2024 | Microsoft Certified: Azure AI Engineer Associate | Certification | [credential] |
| 2024 | AWS Certified Solutions Architect – Associate | Certification | [credential] |
| 2024 | Hugging Face Cookbook — Vision-Language GRPO Recipe | Open Source | [repo] |
| 2023 | PDF Chat — Document Q&A tool via LangChain and HF | Project | [repo] |
| 2023 | George Brown College — Applied A.I. Solution Development | Education | [coursework] |
| 2023 | Deep Learning Specialization — CNNs, RNNs & Optimization | Education | [notebooks] |
| 2023 | Kickstarter Crowdfunding Analysis — SQL Cohort Modeling | Project | [sql] |
| 2021 | MLOps Notes & Reproducible Machine Learning Notebooks | Project | [site] |
| 2021 | Canadian Political Party Email Analysis — Exploratory EDA | Project | [repo] |
| 2021 | Full Stack Open — Modern React, Node & REST APIs | Education | [exercises] |
| — | Carleton University — B.Eng. Electrical Engineering | Education | [degree] |