* Designed and implemented an LLM-driven service to create Alexa Routines, focusing on scalable, configuration-driven code which can handle hundreds of API formats.
* Partnered with science teams to generate supervised fine tuning (SFT) and in-context learning (ICL) training data, evaluate model output, and overcome LLM limitations such as hallucinations and non-determinism. Altogether, increased model accuracy from 58% to 94%, unblocking project launch.
Routine Recommendation Builder
* Redesigned data stores and APIs used to suggest Routines to customers, accelerating recommendation creation, simplifying data contracts, and facilitating personalized recommendations and embedded upsells within Routines.
* Led a team of 5 developers to build an internal website tool for no-code creation of Routine recommendations. Website reduced the creation timeline from 4 weeks to 1 week and enabled project managers to quickly build dozens of recommendations, driving adoption by millions of customers and achieving our yearly engagement goals.
Leadership and Operational
* Managed 2 full-time employees and 1 intern for 6 months and mentored multiple other entry-level engineers and interns.
* Responded to customer-impacting experience-breaking events and addressed service operation issues as part of a cross-team on call rotation.
Work closely with the professor to prepare lectures, and both prepare and grade homework assignments and exams.
Host help sessions open to all students and tutor students individually to explain concepts and help complete homework projects using Java, Python, and C++.
Have worked as TA for: Introduction to Computer Science I & II, Computer Securities, Algorithms
Designed and created a new Recommended Routine presented to all Alexa customers which detects customers’ Smart Home devices and initiates corresponding device actions in real time.
Optimized GUI experience to decrease latency of the Routine homepage within the Alexa app.
Wrote and presented documents detailing the design, architecture, and implementation of a new Recommended Routine. Programmed primarily in Java and Typescript, using React Native framework.