Sep 2020 – Dec 2022 · 2 yrs 3 mos · Remote (from Toronto)

• Working on a project undertaken for J.P. Morgan that involves analyzing legal documents through predictive modeling and information extraction. Some of the tasks taken up are regulation violation prediction, document segmentation through sentence classification, extractive question answering, and overlapping named entity recognition (NER). The methods used range from TF-IDF and SVM to ELMo word embeddings and BERT. A pre-trained BERT model was tuned on domain-specific data in an unsupervised manner and evaluated intrinsically and extrinsically.

Dec 2014 – Aug 2015 · 8 mos

• Worked on the problem of record matching. The existing approach used to identify records across data sets that belong to the same real-world company entity was rule-based. We explored developing a system involving a machine learning-based classifier.

Nov 2011 – Dec 2014 · 3 yrs 1 mo

• Carried out attribute extraction for Bing. The aim of this work was the automatic extraction of attributes such as the address and the phone number from web pages, in an effort to enrich the entity data pool. On more than one market, the extractor was able to obtain data with good precision and recall.

Jul 2009 – Nov 2011 · 2 yrs 4 mos

• Contributed to a camera-based traffic surveillance system, aimed at enhancing the enforcement of traffic laws, that automatically detects a variety of violations live.

Mar 2007 – Jul 2009 · 2 yrs 4 mos

• Worked on the automatic removal of artifacts such as skew and frame in scanned document images.