I harness the capabilities of LLMs by keeping them on just enough leash to be useful

Create a speech-to-semantics system to listen to air-traffic control radio communications, and deliver instructions from ATC to an autonomous aircraft. The system identifies utterances in a live radio feed, determines if an utterance is intended for a given aircraft, and processes the utterance into meaningful commands for a flight planning systemManage a small team of NLP engineers and interns to work on subprojects that improve the capabilities of the above system

Investigate natural language processing problems, including automatic machine translation, semantic parsing, and common-sense reasoning.Build and run state-of-the-art machine translation pipelines that participate in government-sponsored programs such as the NIST LoreHLT task, and the DARPA BOLT and GALE programs.Incorporate ideas from graduate students and professional researchers -- as well as my own published and unpublished work -- into state-of-the-art machine translation and semantic parsing pipelines, and use these pipelines to help students and researchers complete publishable research.Create visualization tools to help researchers target common processing errors, and find possible ways to improve accuracy.Deliver final products used in DARPA and UN field exercises.

Supervise and participate within a small team of programmers, with a goal of improving the accuracy and usability of a speech recognition engine based on CMU SphinxImplement W3C standards for speech recognition and semantic interpretationDevelop APIs and documentation for customersPresent the product to potential customers at trade shows

Lecturer for Mathematics Dept