Research Experience

AI4Ops Research

Applied AI, Comcast
Jan 2022 - Jul 2024

• Led the research and development of log related features in AI4Ops project
• Log mining research to convert text-based application logs to timeseries metric data
• Dependency graph traversal based Root Cause Analysis (RCA) feature over metrics
• LLM based feature integration such as Application Log Summary, Q&A bot, Natural Language to NoSQL DB Query

Attention based Image Captioning System using ResNet and LSTM

Dr. Surekha Bhanot, BITS Pilani
Aug 2019 - Dec 2019

• Designed a system that automatically generated image descriptions
• Produced a BLEU-4 score of 29.1%, which was then state-of-the-art performance on the official Microsoft COCO benchmark
  Keywords: LSTM, ResNet-18, Computer Vision, NLP, Deep Learning
  

Smart Targeted Advertising based on Hyperlocal Factors using IoT and Computer Vision

Dr. Rajneesh Kumar, BITS Pilani
Jan 2020 - May 2020

• Developed an user-centric advertising system leveraging IoT sensors, VGGNet for feature extraction, and AWS for data processing and XGBoost algorithm for accurate ad predictions to recommend ads based on contextual and demographic factors
• Enhanced ad prediction accuracy by ~22% by integrating collaborative filtering into the recommendation algorithm to increase ad relevance and user engagement
  Keywords: IoT, Computer Vision, Recommendation System, Collaborative-filtering
  

Spiking Neural Network Architecture using Magnetic Tunnel Junctions (MTJs)

Dr. Nitin Chaturvedi, BITS Pilani
Aug 2019 - Dec 2019

• Designed a fully connected two-layer neural network with Compound Synaptic Matrix (CMS) and Stochastic Spiking Neuron (SSN)
• Achieved an accuracy of more than 80% for handwritten digital recognition on the MNIST dataset
  Keywords: Neural Networks, Pattern Recognition, Spiking Neural Networks, Magnetic Tunnel Junctions (MTJ)
  

Review paper with implementation of text search engine

Prof. Abhishek BITS, BITS Pilani
Jan 2020 - May 2020

• Authored a term review paper on A. Severyn (Google) & A. Moschitti’s top-tier Information Retrieval conference paper, “Learning to Rank Short Text Pairs with CNNs, SIGIR’15”.
• Subsequently, developed a text search engine for Wikipedia’s corpus data, delivering highly relevant documents for varied queries even with spelling errors & foreign words, evaluation with TREC.