I am an incoming MS CS grad student at University of California, San Diego. I completed my undergrad in Computer Science from Mumbai University, India and have been working on ML/NLP problems for over 3 years now. I am currently working as an applied research scientist at NeuralSpace AI. I am an avid football and Arsenal follower. I also like to read about Indian history and finding out about events and various philosophies that have shaped the world we currently live in. My research interests lie broadly in NLP. I have briefly explained my work and research experience below.
- Accepted and presented at ML-RSA@NeurIPS 2020.
- Presents an exhaustive analysis and evaluation of transformer-based models on Indian languages.
- Analyzes three experimental setups which involve pre-training the language models, fine-tuning multilingual models with data from one language and directly evaluating multilingual models on Indian languages.
- Accepted and presented at IC4S 2019 (http://www.ic4s.org/). Published in Springer series “Advances in Computer, Communication and Computational Sciences”.
- Implemented task-specific LSTM architectures with BERT and XLNet weights using independent and conditional encoding of input text. Also developed a novel architecture that uses Temporal CNNs for stance detection.
- Implemented 3 important papers for the task of Question Answering viz. DrQA, BiDAF, and QANet.
- Each implementation is in the form of a tutorial with detailed explanations about each component/layer.
- Has been starred 165 times on Github.
NeuralSpace (October 2020-Present)- Applied Research Scientist
- I am currently helping the company build an NLP SaaS platform from scratch with focus on low-resource languages. Our platform currently supports 55 languages which includes indic, south-east asian, african, scandinavian and middle-east languages.
- During this, I worked on all the components that are required to build a modern ML/NLP system, right from basic text processing, POS-Tagging to distributed training of models and deploying them at scale.
- Closely worked on natural language understanding (NLU) systems. If your business involves NLU use-cases, reach out to NeuralSpace https://neuralspace.ai for a demo. We are continuously adding a lot of features and have a very exciting plan ahead of us.
NeuralSpace (August 2019-October 2020)- Research Intern
- Developed a document analysis library from scratch. Involves a pipeline that takes in an image and returns all named-entities present in the image. Currently supports template-matching-based entity recognition for any type of PDF document. Also supports the analysis of documents with tabular data.
- Developed a multilingual text summarization model that works for 21 languages including 11 Indian languages.
- Worked on machine translation of low resource languages to English.
Neebal Technologies (June 2018-August 2018)- Software Engineering Intern
- Integrated a handwriting recognition engine into an android app.
- Worked with various Android APIs like BLE, JobScheduler, etc.