Computer science graduate student interested in multilingual natural language processing (NLP), cultural analytics, and ethics and fairness of NLP technologies.
Developed person-aware Natural Language Processing tools, specifically to handle code-switched (mixed-language) text (Adv. Dr. Yulia Tsvetkov, University of Washington)
Main projects:
(1) Developed an adaptable, user-informed model to predict code-switch (language switch) points in mixed English-Spanish dialogues. I presented our work at ACL 2022 in Dublin. In the future we see this classifier implemented into an adaptable dialogue system to control switches between languages.
(2) Using statistical techniques to analyze psycholinguistic drivers of code-switching. There are several existing hypotheses for why people switch languages. Our goal is to identify when each hypothesis is most relevant to categorize different types of code-switch points.
Using stakeholder feedback, iteratively implemented data processing pipelines and dashboards illustrating real-time test performance stats of robotic drive units used in Amazon fulfillment centers (AWS tools: Athena, S3, Lambda, DynamoDB)
•
Employed statistical techniques to identify historical patterns in test data from different drive unit components and to help flag suspicious behavior from new drive units before they are shipped for production
Implemented core services for the company-internal Voice to Text portal for transcribing and analyzing uploaded videos and audio:
•
Developed modular S3 data storage service using Flask. Service could be integrated with any of the team’s AI-driven tools for efficient, reliable, and persistent data storage on Amazon’s S3 cloud.
•
Developed Flask-based orchestration service to integrate Dockerized voice to stext service with S3 data storage service. Used PostgreSQL database to track metadata of uploaded documents and service output results.
•
Experimented with pre-trained voice-to-text model using Kaldi. Researched techniques for unsupervised content classification of transcribed audio to extend the capabilities and usefulness of the voice-to-text service.
