Data Scientist
Current• Built and deployed a web application serving NLP tools using Docker, React, Python on Azure and AWS, with automatically scalable VMs concurrently serving hundreds of users.• Co-developed an STT web application using the same technologies above, currently deployed to on-prem servers for use as part of a Proof-of-Concept with select government agencies, expected to save millions in costs of manual transcription.• Implemented automated software testing using Cypress, including writing… Show more • Built and deployed a web application serving NLP tools using Docker, React, Python on Azure and AWS, with automatically scalable VMs concurrently serving hundreds of users.• Co-developed an STT web application using the same technologies above, currently deployed to on-prem servers for use as part of a Proof-of-Concept with select government agencies, expected to save millions in costs of manual transcription.• Implemented automated software testing using Cypress, including writing component and integration tests.• Trained and tuned Kaldi models with 7,000 hours of data, with a Word Error Rate (WER) of 26% as evaluated on localised and operational context speech data, expected to improve further with the inclusion of 1,500 hours of operational data.• Finetuned Whisper models for STT using Infocomm Media Development Authority’s National Speech Corpus dataset, with a WER of 16% evaluated on the same localised and operational context speech data above.• Integrated Whisper end-to-end into STT web application, including API design and development. Show less