Data Engineer Intern

Overview

TD Bank is a major Canadian financial institution delivering banking, lending, and enterprise technology services across multiple lines of business. As a Data Engineer Intern on the AI Platform team, I worked on production-grade AI data infrastructure supporting document ingestion, embedding generation, RAG retrieval, model endpoint migration, Azure resource management, monitoring, and greenfield AI system design.

Work

• Maintained a production-grade AI ETL pipeline on Azure spanning Azure SQL, Data Lake Storage, Azure Machine Learning, Azure Databricks, and Azure OpenAI.
• Supported ingestion cycles processing 2,500–3,000 unstructured documents, including HTML, PDF, and SharePoint content, across multiple lines of business with distinct business rules and processing configurations.
• Configured and validated ingestion workflows by adjusting pipeline endpoints, verifying chunk-level embedding quality, and executing end-to-end data loads into Azure SQL after each deployment.
• Deployed an enterprise-developed embedding model across the RAG query engine and ingestion pipeline, replacing the Azure OpenAI embedding endpoint.
• Improved retrieval accuracy from 95% to 98% while reducing provisioned throughput from 285 PTUs to 150 PTUs through embedding model migration and infrastructure optimization.
• Executed model endpoint migrations using GitHub Actions CI/CD pipelines and manual VM deployments, selecting the appropriate rollout method based on release state to maintain zero-downtime data ingestion across environments.
• Managed Azure OpenAI PTU and TPM allocation by right-sizing provisioned throughput quotas, optimizing cost-latency tradeoffs, rolling out updated language models, and configuring spillover rules to maintain stability under variable production workloads.
• Built Datadog monitoring dashboards and observability panels across DEV/SIT, PAT, and PROD environments.
• Wrote custom Datadog queries against Azure OpenAI, SQL Database, and VM telemetry to monitor token usage, latency, rate limits, embedding-model memory utilization, SQL DTU consumption, storage availability, ETL defect alerts, service logs, Application Gateway traffic, and VM I/O metrics.
• Established the team's primary source for pipeline health monitoring and incident triage through centralized observability dashboards.
• Designed and configured full infrastructure for a greenfield AI project, defining component topology, connectivity, and resource provisioning from scratch.
• Authored architecture blueprints documenting each component's attributes, dependencies, and deployment requirements to support reproducible infrastructure setup and stakeholder handoffs.
• Placed 2nd out of 19 teams at the TD x Google Gemini Hackathon by designing an agentic multimodal AI copilot using Gemini.
• Built a hackathon prototype that classifies mortgage documents, extracts key data, and performs cross-document reasoning to detect income discrepancies.
• Proposed a workflow reducing conditions review time from 1–2 weeks to under 2 minutes per file across an estimated 1,200 daily renewal files, earning an invitation to pitch to TD's CIO, SVPs, and product owners for potential production deployment.

Stack

Azure SQL, Azure Data Lake Storage, Azure Machine Learning, Azure Databricks, Azure OpenAI, Azure VMs, Application Gateway, Datadog, GitHub Actions, CI/CD, RAG, embedding models, LLMOps, Azure PTU and TPM management, document ingestion, Gemini, multimodal AI agents.