At Uber, I led the development of a production LLM-powered driver-support assistant, from problem framing and evaluation through launch. I designed its hybrid retrieval stack, fine-tuned and aligned the model, and built release checks for groundedness, PII redaction, and out-of-policy responses.
I also built a PySpark marketplace simulator to evaluate pricing policies and applied reinforcement learning to balance rider demand with driver supply, validating launches through switchback experiments. At The Home Depot and Nike, I developed demand forecasts, pricing and recommendation models, and experimentation methods that informed retail decisions.
