Professional experience · Scotiabank
LLM workflows and banking automation
Delivered banking automation using RPA and LLM workflows, with prompt engineering, RAG experiments, and evaluation criteria guiding production readiness.
Product
RPA, agentic workflows, and LLM evaluation
Users
Business users and internal delivery teams
68%
Improvement in response quality
46%
Improvement in task success
76%
Reduction in hallucinations
The problem
The challenge was to translate banking processes into automation requirements that could run within existing infrastructure. I worked across RPA, agentic workflows, and LLM behavior, connecting business tasks with prompt and retrieval requirements, evaluation criteria, and data governance.
My contribution
- Delivered RPA automations and LLM workflows within banking infrastructure, reducing manual processing and contributing to operational expense reductions exceeding $20M.
- Led prompt engineering and RAG experiments across retrieval context and response generation, improving task success by 46% and reducing hallucinations by 76%.
- Defined LLM evaluation criteria for factual accuracy, hallucinations, edge cases, and model drift, improving response quality by 68% and supporting production readiness.
- Specified agentic workflow orchestration, RAG requirements, safety guardrails, and data governance controls for production delivery in a banking environment.
Technical approach
I treated prompt design, retrieved context, and generated responses as parts of the same workflow. Evaluation criteria covered factual accuracy, hallucinations, edge cases, and drift, while task success connected those technical measures to whether the automation helped someone complete their work.
This summary describes my scope and outcomes. It omits internal prompts, datasets, architecture, and customer information.