Mar 2021 – Jun 2022 · 1 yr 3 mos · Bengaluru, Karnataka, India

• Lead planning and development of data science use-cases and deployed in production environment to provide personalised experience for ~10 Mn users.
• Devised in-house ML framework to solve user cold-start problem in personalisation algorithms.
• Increased Day 0 retention and watch time in production environment (~80% users) after running 18+ experiments in A/B test environment.
• Designed content seeder framework to solve item cold-start problem using OpenAI CLIP and Milvus (vector search database).
• Reduced median number of days to reach 100 views for a cold/new video using GBDT models, modified sigmoid form equations in recommender systems.
• Responsible for extracting insights from user data to understand any anomalies, trends, user behaviour.
• Incorporated user generated content videos in personalisation feed using description based text and image embeddings.
• Built robust performance metrics framework to run A/B tests and measure true potential of various
• personalisation experiments across the teams.
• Developed various positive engagement event classification rankers using GBDT, Neural Networks to rank multiple candidate sets and provide more engaging/meaningful video feed to users across vernacular languages (~12).
• Mentored junior data scientists and Interns.

Jun 2018 – Mar 2021 · 2 yrs 9 mos · Mumbai Metropolitan Region

• Collaborated with multiple Clients to leverage AI/ML based products
• Lead planning and development of data science use-cases and reported improvements in product features to stakeholders
• Developed multiple ML algorithms for piracy detection of softwares for fortune 500 client and generated leads accounting for 50% of annual revenue of License & Compliance Team
• Automated machine learning life-cycle for piracy detection system using Jenkins and Oozie enabling evidence-based leads generation on a month-on-month basis
• Devised PCA based unsupervised piracy attribution score by creating interpretable and manipulable demographic profiles of software usage data for costumers to prioritize generated leads
• Designed and developed supervised regression models to quantify piracy in a region/country by analyzing real time usage data.
• Improved existing association based estimates with more robust and precise predictions of piracy.
• Devised an automated methodology for company name standardization using NLP and fuzzy-logic based techniques in PySpark
• Continuously monitor and anticipate project needs using JIRA and make proactive recommendations week-on-week following agile methodology and accomplishing near 100% task resolution
• NLP
• Implemented multiplicative LSTM language model based on paper ”Learning to Generate Reviews and Discovering Sentiment” using Keras
• Developed unsupervised sentiment analysis pipeline for costumer reviews which later was integrated with Fractal’s AI platform - IVA
• Analyzed performance of unsupervised sentiment prediction model on ”Twitter Sentiment Analysis”(Kaggle) competition and achieved F1 score of 0.77

May 2017 – Jul 2017 · 2 mos · Whitefield,Bangalore

• Designed automated model on Computed Tomography for Headrest Removal using Tensorflow
• Image segmentation of Head CT scan dicom images using convolutional neural networks
• Classified faulty products in CT machine by developing Text-Based Error Classification model
• Github link: https://github.com/harsh1795/CNN-DICOM-Segmentation

May 2016 – Jul 2016 · 2 mos · 10, Masjid Moth Greater Kailash 2 New Delhi, Delhi 110048 India

• Data Analyst
• Analysis on Indian Movies
• Gender analysis on Indian movie scripts
• Optimized tour plan for Indian Tourism
• Github link: https://github.com/harsh1795/TechMonkey

Mar 2016 – Jul 2016 · 4 mos · Greater Chennai Area

• Working on a project involving Recommendation Engines. We are using the movie-lens database to develop a movie Recommendation engine that uses collaborative filtering and alternating least squares. It is currently being built on python, with the focus being on switching to Apace-Spark in the coming weeks.
• Github link : https://github.com/rsk2327/RecommendationEngine

Nov 2015 – Feb 2016 · 3 mos · Greater Chennai Area

• Formulated Irwin-Hall’s distribution for maximization of profit for a firm’s product using Python2
• Plotted behavior of decision variables like product price, informative and comparative advertising money with profit using Matplotlib

May 2015 – Jun 2015 · 1 mo · Chittorgarh (Raj.)

• Analysed operations of Cell House in Zinc Hydrometallurgy Plant which produces Zn(99.95%) by electrolysis process
• Proposed Analysis Report of Zinc Production process depending on Process parameters using Excel
• Presented various effects of Impurities on Zinc Sheet