About

Rohan Purswani is an experienced ML engineer with a demonstrated history of designing and developing ML-based features for voice assistants like Amazon Alexa and Samsung Bixby. He is currently working as an SDE-II, ML at Amazon Search, where he is designing and developing micro-services for ingesting end-user Alexa synced contacts in the cloud and making them voice searchable. He is also employing deep neural networks and other statistical ML techniques for runtime connection resolution and contact disambiguation while placing an Alexa call. Prior to his current role, Rohan worked at Samsung as a Product Manager, where he led the PM activities for the ASR module globally on Bixby, Samsung’s voice-powered virtual assistant for flagship phones. He enhanced solution competence by identifying real-world scenarios where speech recognition needs to happen on a smartphone, influencing data sourcing, training and analysis strategies, and finally defining data governance guidelines, tech advancements, and revamping KPIs. Rohan also established processes to onboard real-time training data, partner data, and end-user production logs for language model enhancement, improving Word and Sentence Recognition rates relatively by 10%. Before his stint as a Product Manager, Rohan worked as a Speech and NLP Senior Engineer at Samsung, where he developed an end-to-end on-device ASR system for Bixby, Samsung’s virtual assistant, and deployed it for dictation on flagship smartphones in August 2019, obviating network connectivity and reducing latency. He employed multi-task learning and sequential transfer learning based strategies for online attention-based encoder-decoder models, improving the performance by 10%. Rohan also led the statistical language model development for Bixby ASR, accented for US, British, and Indian English. Rohan Purswani holds a Bachelor’s Degree in Information Technology from Veermata Jijabai Technological Institute (VJTI), where he studied from 2013 to 2017. He also completed his All India Senior School Certificate Examination in Science from Birla School, Kalyan, in 2013, and his High School from the same institution in 2011. Rohan's tech stack includes Software Engineering, ML, Search, Product Management, Language Models, Neural Networks, Microservices, Mobile, and NLP. He has 5.71 years of relevant experience.