Mar 2023 – Nov 2024 · 1 yr 8 mos · Remote

• recommendation engine pipeline

May 2021 – Apr 2022 · 11 mos

Sep 2019 – May 2021 · 1 yr 8 mos · Gurugram, Haryana, India

Aug 2014 – Aug 2019 · 5 yrs · Noida, Uttar Pradesh, India

• User retention optimisation engine :
• Enhanced user retention by 16% on an average, leading to 2X DAU numbers for the gaming app client, policy promotion optimization for banking client
• Data analysis and pre-processing insights extraction around pandas (dataframe)
• Implementation of Random Forest Algorithm to determine features/insights ranking and corroborate finding b/w 2 approaches
• Image Processing
• Deep Learning (Artificial Neural Networks)
• Classification of home appliances using Neural Network
• Trained a model from scratch for classification of home appliances such as microwave, refrigerator, digital camera, phone/tablet, computer/laptop, keyboard, printer, speaker etc.
• Age and Gender Classification model training and compressing from scratch
• Gender and Age classification model creation using neural network (deep learning) from scratch
• Compression of Neural Network model for porting to small devices
• Neural Network classification models are usually of order of 50-500 MBs, using pruning of weights and further clustering weights in chunks, model size was reduced to order of ~600 KBs without losing accuracy of the classification
• Image Quality for classification of Blurred Images
• Used Pre-trained Alexnet architecture to extract features from penultimate layer ( before Softmax layer) and subsequently trained these features on different kernels of SVM to get optimum results
• Improved performance by implementing features extraction through VGG-16 network and Restnet-50 network and subsequently fine-tuning these models by adding fully connected layers
• Churn Prediction and Insights
• Prediction of user uninstallation based on features based on user's activity, usage pattern, app crash, phone storage
• Used random forest classifier to train model for the purpose of this classification

Aug 2014 – Mar 2015 · 7 mos · Noida, Uttar Pradesh, India

• Text Analysis:
• POC of chatbot based on sentence paraphrasing (influenced from Kaggle quora-question-pairs approach), using contemporary approaches based on RNN and manual features extraction followed by training a classifier on these features.
• Implemented Naive Bayes to classify messages (SMSs) between commercial and non-commercial. Extraction of NER from commercial SMS data to extract information such as name, place, merchant name. Further trained classification model on names (name, place, merchant) using Maximum Entropy classifier (Maxent)
• Creating module for users competing apps information and usage comparison
• Module creation for extracting users' interest faced on usage and apps installation history
• Automation of demo account for encompassing every information to be shown on an apps dashboard
• Data scrapping through web time and again using beautiful soup module in python
• Monitoring scripts for data sanity
• Elementary Knowledge:
• Mongo query optimization for faster update and data fetching in related projects
• Implementation of lambda function in node.js
• Mongo storage architecture optimization for swift updates on dashboard
• Data pipe line creation using lambda architecture for batch and stream analysis of data. Worked on streaming data analysis using Apache Storm