Building production AI at scale — agentic systems, RAG, and ML pipelines
Machine Learning Engineer & Agentic AI Engineer at Loblaw Digital · Toronto, ON
I architect enterprise agentic AI, production RAG, and shared platform infrastructure — shipping intelligent systems from design to deployment at retail scale.
Previously at Priceline (Booking Holdings), I built Vertex AI ML pipelines, a scalable GenAI framework with LLM-as-a-Judge validation, and NLP features including hotel review summarization that drove a 30% conversion uplift in A/B testing.
M.Sc. Computer Science — University of Alberta · 6 peer-reviewed publications
| 5+ Years | 97% RAG Hit Rate | $75K+ Cost Savings | 30% Conversion Uplift |
|---|---|---|---|
| 6 Publications | MCP Agent Platform | Vertex AI Pipelines | $10M+ Client Impact |
97% retrieval hit rate — I still know where the other 3% live.
- Agentic platform architecture — planner-based agents, guardrails, and multi-tier memory
- Agent Tools Package — centralized MCP tooling adopted across AI product teams
- Multi-agent orchestration — evaluation frameworks and production deployment patterns
Most of my week is evals, logs, and the occasional "why did it do that?"
| Area | Impact |
|---|---|
| Agentic AI | Led planner-based platform architecture with guardrails and long/short/episodic memory |
| Production RAG | Hybrid-search system — 97% hit rate, 78% precision, 97% recall, $75K–$85K projected savings |
| GenAI Framework | Scalable GCP framework with LLM-as-a-Judge validation, rate limiting, and CI/CD |
| ML Pipelines | Standardized Vertex AI pipelines with feature stores, versioning, and Airflow orchestration |
| NLP at Scale | Hotel review summarization with 30% conversion uplift; MIRA chatbot with 97% intent F1 |
Agentic architecture at a glance
graph LR
Q[User Query] --> P[Planner]
P --> R[Retrieval / RAG]
P --> A[Specialized Agents]
A --> M[Memory]
A --> T[MCP Tools]
P --> G[Guardrails]
AI / Agents
Cloud / MLOps
|
NLP · SemEval 2020 Commonsense validation & explanation with RoBERTa, BERT, and GPT-2 on SemEval Task 4 benchmark data.
|
|
Recommendation Systems · Deep Learning Personalized music recommendations using content-based filtering, clustering, and deep learning on Spotify data.
|
|
Time Series · Forecasting Flight price forecasting with ARIMA, Prophet, XGBoost, and LSTM — aligned with production travel pricing use cases.
|
|
|
| Year | Title | Venue |
|---|---|---|
| 2023 | Intent and Entity Detection with Data Augmentation for a Mental Health Chatbot | ACM IVA |
| 2022 | AI-Guided Mental Health Chatbot for Health Care Workers | JMIR Research Protocols |
| 2022 | Developing a Mental Health Virtual Assistant for Healthcare Workers | M.Sc. Thesis, UAlberta |
| 2021 | Efficient Load Balancing for Service Function Chain Mapping | Computers & Electrical Engineering |
Open to collaborations on agentic AI, RAG, and production ML systems