I bridge applied AI research, production software, and stakeholder-facing delivery.
I build useful AI systems where the hard part is not just the model, but the connective tissue around it: messy data, retrieval, cloud infrastructure, evaluation, safety constraints, stakeholder trust, and deployment reality.
My work spans secure government analytics at Statistics Canada, production-grade R and Python migration, OCR and reinforcement-learning research for public-sector partners, client-facing generative AI workflows, local-first LLM apps, RAG engines, and GCP-based agent prototypes.
Co-leading the migration from legacy SAS to open-source R and Python in a secure production environment. I build validation pipelines, internal analytics tools, peer-review workflows, and GitLab practices that help teams ship reliable statistical systems with less manual drag.
Built applied ML systems for public-sector partners, including OCR pipelines for Health Canada and reinforcement-learning simulations for PHAC. Led survey release workflows, founded StatCan's first Kaggle Competitive Group, and published research on food insecurity and public health modelling.
Built client-facing GenAI pipelines, local LLM applications, RAG systems, GCP/Vertex AI agent prototypes, and AI safety guardrails. I like the whole path from ambiguous problem to working system: discovery, architecture, code, evaluation, debugging, and handoff.
Reinforcement learning, public-sector data science, and interpretable modelling
Agent simulation, public health, and decision support
Co-developed an agent-based reinforcement-learning simulation using Ontario population data to study COVID-19 mitigation strategies. The work connected ML experimentation, epidemiological modelling, dashboarding, and policy-facing research.
Statistical research, survey data, and social impact
Published Statistics Canada research on stressful life events and food insecurity, grounding technical analysis in real social outcomes. This work sharpened how I think about evidence quality, sensitive data, and responsible model interpretation.
Client-facing AI systems, rapid prototyping, and creative deployment
From open-source research to shipped client deliverables
Built custom generative AI workflows for commercial campaigns and music videos, translating unstable open-source tools into repeatable pipelines under real deadlines. This is the same operating mode I bring to applied AI engineering: prototype quickly, debug deeply, and ship something usable.
Agentic systems, RAG pipelines, local LLMs, AI safety, and applied machine learning
Python, LangChain, Qdrant, FastEmbed, Ollama, AWS S3, Docker
Built a local-first retrieval-augmented generation system for 20GB+ PDF libraries, with resumable ingestion, multiprocessing, size guards, vector indexing, MMR retrieval, source-attributed answers, and operational health/audit scripts.
RAG · vector databases · unstructured data · retrieval evaluation · local LLMs · observability
Flutter, Dart, Gemma, LiteRT-LM, SQLCipher, local evaluation
Designed a local-first AI health tracking app for IBD pattern review. Deterministic code owns risk scoring and persistence, while Gemma generates grounded explanations from local evidence without medication advice or unsafe automation.
on-device AI · privacy-preserving ML · safety boundaries · mobile AI · local inference
Mila AI for Good Hackathon, Special Recognition
Helped build a youth mental-health AI safety system with a six-phase synthetic data generation pipeline, curriculum-learning strategy, classifier guardrails, validation workflows, and evaluation scripts for safer conversational AI.
AI safety · synthetic data · curriculum learning · guardrails · classifier evaluation · mental health AI
Repository cannot be public because of hackathon and safety constraints; implementation details can be shared privately.
In Progress · GCP, Vertex AI/Gemini, Plaid, Twilio, Firestore, Terraform
Building an SMS-first personal finance agent that syncs bank transactions through Plaid, classifies spending with Gemini on Vertex AI, and uses event-driven GCP services for budget workflows and transaction state management.
agentic workflows · GCP · Vertex AI · Cloud Run · Cloud Functions · Pub/Sub · state management
In Progress · Multi-agent systems and tool orchestration
Prototyping agent infrastructure for tool use, task routing, retrieval, and repeatable local/cloud workflows. Focus areas include MCP-style integration, tracing, prompt contracts, and production-minded debugging.
multi-agent systems · MCP · tool calling · ReAct patterns · tracing · prompt engineering
In Progress · Hugging Face AI Agents Course, DeepLearning.AI Generative AI with LLMs
Strengthening the theory and implementation patterns behind production GenAI: agent design, evaluation, LLM application architecture, transformers, fine-tuning concepts, and responsible deployment practices.
LLM applications · agents · transformers · evaluation · Hugging Face · GenAI systems
Earlier ML and statistical learning work
Supervised and Unsupervised Machine Learning
Compared supervised and unsupervised learning algorithms for predicting heart disease outcomes, with emphasis on model selection, feature behavior, and interpretable evaluation.
classification · clustering · model evaluation · healthcare analytics
Digital Forensics Research
Researched steganalysis algorithms for detecting hidden information in digital media, including technical discussion of statistical signals, image features, and digital forensics applications.
statistical detection · digital forensics · image analysis · security
Interested in GenAI engineering, forward-deployed AI, data science, RAG, agents, or cloud AI systems? Connect with me below.