scotiabank - ml engineer intern sept 2025 - may 2026
built an llm-as-a-judge eval and red-teaming framework (python, langchain, vertex ai). in-house eval cycle went from over a week to 50 minutes across 1000+ examples. (this is not a typo)
also: spark + airflow etl pipelines on gcp (60% faster processing), and a powerbi/excel dax to bigquery migration (40% faster loads). the data team sleeps better now.