Working on Transformers and Large Language Models(LLMs)

Working on creating an end-to-end pipeline for labeling and classification of different types of documents from BPO agents. Worked from Extraction, preprocessing, and training to Inference for different pipelines in production.

●Worked on creating a service that can understand the conversation of 2 people using Semantic Analysis. Topic Identification(triggers, emotions), Concept Extraction, Dependency parsing, etc.●Worked on Flight Booking Recommendations for a Big Airlines Client for multiple use cases, like Overbooking, Spillage, and Spoilage.● Worked on Patient Switch prediction using multiple techniques like Positive Unlabeled Learning, Classification, Regression, etc.

Worked on parsing order receipts and bills using different text parsers. NLP and computer vision. Also worked on Fraud detection of duplicates bills using computer vision techniques

♦A comprehensive review of the literature was done to determine the best approaches to training data set development as well as the best categories to use for labeled.♦Dataset was developed to represent urban structural types using optical remote sensing data (Sentinel-2)that is freely available and labeled using land use planning and census data. Dataset is created using Sentinal 2 for Argentina and Kenya cities using ArcGIS.♦A model was developed using the Convolution Neural Network(CNN) from scratch to predict slums in the cities.♦ Due to the highly imbalanced dataset, used the oversampling technique.♦ Computed per class accuracy and used confusion matrix, F1-score to evaluate the result.Lab: GeoLead lab(Envi Lab)Guide: Prof. Christopher Storie