SDE with Amazon Connect, Analytics team

- Reverse-engineered large language models (GPT2) to study specific model behaviors using mechanistic interpretability.- Worked on interpreting unexpected/expected memorization of text sequences in GPT2-small.- Analyzed the direct and indirect effects of various layers on the model's prediction to understand the behavior at the neuron level.

TA for CSE 291: Unsupervised Learning with Prof Sanjoy Dasgupta (Spring 2023)TA for CSE 158: Recommender Systems and Text Mining with Prof Julian McAuley (Fall 2022)TA for CSE 256: Statistical Natural Language Processing with Prof Ndapa Nakashole (Spring 2022)

- Worked at the intersection of computational neuroscience and NLP under Professor Maxim Bazhenov.- Applied the idea of biological sleep to artificial neural networks to test if a sleep phase helps with generalization.- Ran experiments under two settings: continual learning and robustness to noisy inputs in NLP.

- Developed a cloud-based SaaS platform that enables developers, practitioners, and managers to leverage the power of Natural Language Processing (NLP) inside their organizations without any data science/NLP knowledge.- Worked on the entire pipeline from basic text processing to distributed training and deploying NLU models at scale. - Scaled the platform to support 55 languages which include a number of low-resource languages spoken by millions of people in India, the middle-east… Show more - Developed a cloud-based SaaS platform that enables developers, practitioners, and managers to leverage the power of Natural Language Processing (NLP) inside their organizations without any data science/NLP knowledge.- Worked on the entire pipeline from basic text processing to distributed training and deploying NLU models at scale. - Scaled the platform to support 55 languages which include a number of low-resource languages spoken by millions of people in India, the middle-east and Africa. Show less

- Built an extractive text-summarization model using sub-word semantic hashing to encode sentences as sparse vectors. The model supports 21 languages including all major Indian languages.- Developed a document analysis library that processes PDF and JPEG files and returns all text entities. The library pipeline involves edge detection, perspective transformation, template alignment, and OCR.- Added support for template-matching-based entity recognition of any PDF document including… Show more - Built an extractive text-summarization model using sub-word semantic hashing to encode sentences as sparse vectors. The model supports 21 languages including all major Indian languages.- Developed a document analysis library that processes PDF and JPEG files and returns all text entities. The library pipeline involves edge detection, perspective transformation, template alignment, and OCR.- Added support for template-matching-based entity recognition of any PDF document including analysis of documents with tabular data. Show less

Created assignments for sophomores to help them get used to various tools and technologies. Guided them to create an inventory management application. Held meetings to discuss features and challenges.

As a junior year student mentored sophomores to build an attendance application that allows teachers to collect attendance on their mobile phones and also allows students to keep track of their lectures and attendance. The app has been published on Google Play Store and is used by teachers and students alike.

Developed multiple RESTful APIs and integrated them with the android app built for virtual learning that enables teachers and students to upload/download assignments, notes, and conduct quizzes.

- Worked on a client application that aims to digitize the transactions that occur between a doctor and his patients. The app interfaces with a smart prescription paper and smart pen which digitizes your handwriting instantly as you write onto your prescription paper. - Integrated the handwriting recognition engine into the app and worked on multiple other app components.- Facilitated smoother user experience by using the JobScheduler API to manage concurrent processes and improve… Show more - Worked on a client application that aims to digitize the transactions that occur between a doctor and his patients. The app interfaces with a smart prescription paper and smart pen which digitizes your handwriting instantly as you write onto your prescription paper. - Integrated the handwriting recognition engine into the app and worked on multiple other app components.- Facilitated smoother user experience by using the JobScheduler API to manage concurrent processes and improve memory allocation within the app. Show less