May 2019 – Jul 2019 · 2 mos · New York, USA

• Worked on detection of scene boundaries in video episodes, to be used for placing ads in video streaming and predicting topics in a video.
• Formulated the task as a classification problem. Experimented with loss function and different ways of input data creation to tackle data imbalance.
• Proposed and implemented a BERT based model which converts chunks of captions into equivalent representation and tests whether they are from the same scene.
• Designed a local maxima based approach to decrease false positives while testing in sliding window fashion on episodes.

May 2018 – Jul 2018 · 2 mos · Bengaluru, Karnataka, India

• Built models for constructing future Repurchase Agreement(Repo) rates for all the GC sectors of EUR and GBP currencies.
• Improved over existing model to take into account for rate changes due to IMM rolls and events like quarter-ends, year-ends etc.
• Built an end-to-end pipeline for taking live data from traders using GUI, save it to time-series databases at the End of Day (EOD) and use this data to make curves.

May 2017 – Jul 2017 · 2 mos · Greater Lucknow Area

• Designed various intuitive gestures for day-to-day use in laptops. Created dataset for this in fluorescent and day-lighting conditions using a webcam.
• Proposed a method to extract features from frames using contour information of hand. Trained a classification model on features vector concatenated in time for 25 frames.
• The model achieved an accuracy of 94% on a dataset containing approximately equal no. of sequences from both lighting conditions.