Jul 2025 – Present · 8 mos

• • Multilingual Speech LLMs, Codeswitching, Language Identification

Jul 2023 – Jun 2025 · 1 yr 11 mos

• Core contributor for the research and training of DuRep - a state-of-the-art multilingual speech encoder that powers Amazon's Nova Sonic Speech-to-Speech LLMs
• First-authored a paper on the DuRep encoder, presented at Interspeech 2025, Rotterdam
• Researched and trained a unified speech-audio encoder designed to generalise across diverse speech and audio tasks
• Trained and evaluated models for Alexa's enIN ASR system, identifying and resolving critical issues

Aug 2022 – Jun 2023 · 10 mos

• Part of Automatic Speech Recognition team at Alexa Speech
• Implemented and experimented with Adversarial networks, Multi-Task models and Mixture of Experts based models (MoEs) for Multilingual Speech Recognition
• Pretrained a 1 Billion+ Multilingual Speech Foundation Model using multi-stage approach on Alexa data