At Robinhood, I build the LLM systems behind a customer-facing support assistant and an internal care-agent tool. They now resolve a majority of inbound support contacts without agent involvement.
I designed retrieval over help content, brokerage policy, and live account context, with re-indexing as policies change. Moving to hybrid retrieval with reranking raised context precision roughly 30% on a held-out query set.
I implemented authorization-scoped tool calls and the guardrail and PII-redaction layer for requests. I also built evaluation and tracing systems that measure retrieval quality, response accuracy, and tool-call failures.
Before that, at Robinhood I worked on observability for trading, payments, and options services. At AWS, I productionized risk models and built real-time scoring infrastructure; earlier, I developed licensing and billing services at Microsoft and trading systems at Trading Technologies.
