Sep 2023 – Feb 2024 · 5 mos

Dec 2021 – Aug 2023 · 1 yr 8 mos · London, England, United Kingdom

• At Meta, I led post-Ad-click full funnel optimisation in Shops vertical to maximise expected GMV to maximise ROAS. I was part of exponential growth trajectory crossing 1B USD revenue from Shops vertical during my tenure. I was tech lead for the post click dynamic ranking workstream comprising efforts from around 10 teams and involving problems like product ranking, creating pivots, pivots ranking, whole page ranking, image ranking, features authoring etc quite a few shops surfaces on Instagram and Facebook. Our efforts drove impact of more than 10% ROAS for advertisers. We have used state of the art deep learning and reinforcement learning approaches to serve billions of users.

Sep 2015 – Nov 2021 · 6 yrs 2 mos · Bangalore

• At Embibe, I was on a mission to build an AI platform for delivering learning outcomes at scale. I founded, built, and led the Data Science team of 50+ data scientists, ML engineers, and data engineers. Additionally, I founded Ember Research Lab to streamline efforts, resulting in 15+ publications and patents for research initiatives involving learning outcomes, adaptive learning, automated test generation, academic knowledge base, content generation, knowledge-infused deep learning, and interpretability and explainability of AI models. I also led the IP diligence, roadmap, and execution for Jio’s 180M USD acquisition of Embibe.

Apr 2013 – Sep 2015 · 2 yrs 5 mos · Bengaluru Area, India

• At McKinsey, I was the AI/ML leader for the new initiative, McKinsey Digital Labs, within the firm. I provided technical leadership on high-impact problems such as optimizing network utilization for railroads, constrained scheduling in industrial plants, social graph insights for large organizations, sales representative allocation, and more for various clients. One such high-impact project delivered a $400 million impact for the largest railroad in the USA by implementing optimized network utilization solutions. These solutions involved optimizing train configurations, locomotive scheduling, routing, demand forecasting, and crew scheduling while adhering to given SLAs.

Nov 2011 – Mar 2013 · 1 yr 4 mos · Bangalore

• My role at Runa (HOS) is to handle predictive modeling and machine learning aspect of company's products - Instant PerfectOffer and PerfectShipping.
• Instant PerfectOffer is a real-time, data-driven automated offer that maximizes sales conversion while minimizing margin impact. It involves extracting in-session features, aggregating historical features from time series data stream, and deriving predictive models to predict purchase intent and offer elasticity.
• Perfect Shipping predicts delivery time for shipment of product purchased online, i.e. ebay listings. It spans from modeling sellers behavior, postal service's behavior till deriving correlations due to other factors like holidays, shopping season, natural calamities etc
• Tags: decision trees, support vector machines, logistic regression, gamma distribution, bayesian, time series analysis, maximization, dynamic pricing, real time predictive analysis, unsupervised clustering, big data, cloud computing

Jul 2006 – Nov 2011 · 5 yrs 4 mos · Bangalore

• I was involved in Information Retrieval, Data Mining, and Machine Learning @ Veveo. It spans from developing backend knowledge based systems till serving personalized search/recommendation results by interleaving responses from many vertical search engines using content preferences learned about a user.
• Few of the projects I worked at Veveo are,
• Personalized Federated Search - It submits search request to many search engines of vertical data-spaces, i.e. universal smart tags, local listings, points of interests etc. Interleaves search results therefrom customized using content preferences learned about a user
• Universal Smart Tagging - It mines unstructured data from web to create and deduplicate smart tags (i.e. personalities, movie, album, song, crew, concept, points of interests). Classify around few hundreds of millions of crawled videos, news articles, text segments to one or many such smart tags.
• Phrase Extraction and Relevance Computation - Extracts multi word phrase from given text segment, Assigns relevance based on global relevance as well as local semantic importance of phrase using universal smart tagging.
• Tags: classification, tf-idf, bayesian, local semantic analysis, global relevance computation, consine similarity, natural language processing, support vector machines, template based parsing, pagerank, deduplication, data enrichment, supervised classification, sparse matrix, associative rule mining, metadata augmentation, asynchronous network io, serialization and deserialization, memory footprints optimization, intersections, disjunctions, personalization