May 2022 – Aug 2022 · 3 mos · New York, New York, United States

• Systematic Quant Hedge Fund
• Received return offer to join the firm as a full time Quant Researcher.

Jul 2018 – Jul 2021 · 3 yrs · Greater Bengaluru Area

• Quantitative Strategist(Strats)

Aug 2017 – May 2018 · 9 mos

• Building chatbots for enabling e-commerce for quick service restaurants and implemented chatbot framework for handling various use cases on e-commerce sites. Developed an entropy minimizing model that reduced the number of preference elucidation steps required from the user to allow for a smooth order placement. Also built a chatbot for amazon(POC Sample) to allow user for efficient order placement, Q/A mapping to queries, review filtering and integrated a collaborative filtering based recommendation system.
• This was part of joint research program between Capillary Technologies and Computer Science Department, IIT Kharagpur. (Center of excellence in AI)

May 2017 – Jul 2017 · 2 mos

• Received return offer to join the firm for a full-time position*
• Developed models to extract the store-identifier and geo-tagging information from the description field of huge proprietary credit card transactions datasets to calculate a "popular metric" used by portfolio managers to forecast the sales and performance of a retail chain company.
• This project involved fuzzy parsing and NLP applied to unstructured raw data with several billion rows. Parallelized code to allow scalable execution of the pipeline over large datasets.
• Achieved an accuracy of over 94%(for a set of tickers)along with confidence tags to reduce manual intervention drastically. Achieved geo-location identification with a 60% better extraction rate than the vendor data tagging.

May 2016 – Jul 2016 · 2 mos

• Used OpenCV and Sklearn to design algorithms which efficiently(over 95% accuracy) name all the major colours in all kinds of apparels by using KMeans on LAB space and assigning major chunks. Designed a merging strategy where every pixel rectifies the surrounding pixel where there is a conflict of assignment(as color space is not spherical and euclidean metric doesn't apply).
• Designed heuristic based pattern matching algorithms to classify apparels into 5 major categories with an accuracy of 80% and reduced manual tagging requirement drastically, by assigning confidence tags.
• Used transfer learning to retrain the last layers of googlenet based model previously trained on Imagenet to classify shoes into 12 different categories. Achieved an accuracy of ~99% (best case for completely different looking shoes) and worst case accuracy of ~90% (shoes which are difficult to resolve) over binary classification between two categories. Used binary classification to classify a footwear in multiple steps to its final class. Used Caffe framework and automated training, validation and testing process by scraping images from e-commerce site and built an API, which just requires the class names to be separated.