I build AI systems for workflows where mistakes are expensive, context is messy, and users need to trust the output before they can act on it.

At TP ICAP, I work on an FX options desk and build LLM-powered tools around the workflows brokers use every day. That includes a platform that translates client directives into desk-native language, routing logic for complex option structures that reduced manual ticket creation by 30%, and a trade-origination system that turns public post-trade data into ranked ideas brokers use in live client conversations.

I like building the full loop from the messy source data and model behavior to the workflow, user feedback, and product decisions that determine whether a system actually gets adopted.

Outside work, I build AI-native tools for myself. My reading archive uses retrieval, generated quizzes, adversarial summaries, and source-grounded chat to help me understand and retain what I read rather than just saving it. I am interested in LLMs, AI product design, and deploying software people actually rely on.