May 2017 – Aug 2017 · 3 mos

• I worked with the organization NumFOCUS, to integrate the topic-modelling library Gensim with scikit-learn and Keras by creating wrappers around all Gensim models. The project simplified Gensim’s API and allowed seamless integration of Gensim with these libraries for supervised learning. I also simultaneously collaborated with the library 'shorttext' by updating code which used Gensim models.
• As part of my GSoC project, I also added a Python implementation of Facebook Research's fastText model to Gensim. FastText is a model for efficient learning of word representations and achieves state-of-the-art results for several tasks which involve the user of word-embeddings.

May 2017 – Jul 2017 · 2 mos · Greater Vancouver Metropolitan Area

• I worked on the project 'Wikipedia Text Processor' by mining English Wikipedia in its entirety to extract and model historical text. The developed system learned to represent historical data and classify new unseen text as 'historical' or 'not historical'. The system could also be generalized to model categories of text based on any idea of 'interestingness' input by the user.

Dec 2016 – Jan 2017 · 1 mo

• I developed a pregnancy health assistant for Facebook Messenger to provide localized and doctor-verified information to users. The information shared with the user was based on the particular user’s current week of pregnancy and previous interaction with the assistant. I also developed an assistant for Google Home using Actions on Google and api.ai.

May 2016 – Jul 2016 · 2 mos · Mumbai Metropolitan Region

• I worked on a system to allow users to independently modify trade and market settings and visualize the changes in risk values as a result of these modifications. The project allowed users to compare risk values for the actual and modified markets corresponding to Nomura’s daily trades. The developed system was to be used daily across all offices of Nomura globally and was deployed in my last week of the internship.
• I also implemented bond convexity calculations for certain market settings.

May 2015 – Jun 2015 · 1 mo · Greater Hyderabad Area

• I implemented an open-domain question-answering system for single-relation questions. The system was based on a paper by Microsoft Research titled 'Semantic Parsing for Single Relation Question Answering', which introduces a novel semantic parsing framework for the task.