At Krisp, I designed and developed end-to-end training, inference, and evaluation pipelines for an Armenian automatic speech recognition project. I also benchmarked open-source speech models and designed lightweight on-device and high-capacity server-side deployment variants for Armenian-English code-switching.
At EPFL’s Laboratory 4 (LTS4), I researched discrete flow matching models for graph generation and how sampling-time controls could improve generation quality. I was selected for EPFL’s E3 Program in Signal Processing.
My projects include a Flask-based concert-tour question-answering service using retrieval-augmented generation, OpenAI, and ChromaDB. I also contributed frontend and AI/data science work to Bookfinder, an AI-powered book discovery platform for the Yerevan market.
I earned a Bachelor of Science in Data Science from the American University of Armenia, where I was the program’s top student for four consecutive years. My work and studies span speech processing, machine learning, and statistical methods.
