Rohit Rajkumar Dhonukshe

Designed and executed a control experiment comparing AI-recommended vs. manually created Ad campaigns. Applied Fixed-effect Regression, Marketing Mix Modelling (MMM) to analyze performance data, revealing superior results from manual campaigns, demonstrating the value of sellers’ private information in campaign optimization. • Spearheaded the Quantitative analysis on the impact of the Promoted (Sponsored Search) Ads channel on seller performance using R, utilizing Causal Inference technique - Propensity Score Matching to identify adopter, non-adopter seller pairs of the new Ads channel. • Conducted hypothesis, A/B testing, Average Treatment Effect on diverse seller data (performance history, product portfolio), uncovering statistically significant improvements in net sales, clicks for channel adopting sellers in 2023. • Implemented Staggered Difference-in-Difference model to predict Sales Uplift of Ads channel adoption, by determining Conditional Average Treatment Effects of Sponsored Search on seller performance, enabling targeted recommendations for high-potential sellers. • Conducted PCA on items’ leaf category features to derive principal components, enabling more granular control in the comparison of AI-recommended and manually created ad campaigns, further validating the superior performance of manual campaigns. • Developed Causal Trees, Causal Forest (Generalized Random Forest) model to determine the effect of match types (exact, phrase, broad) on clicks, revenue, identifying seller segments, item categories, keyword groups where specific match types drive optimal performance. • Developed a specialized Data Mart within Hadoop ecosystem (HDFS) using SQL, Spark, ETL processes, significantly accelerating analytical research by providing streamlined access to relevant datasets. • Investigated seller dynamics of the advertising channel by visualising trends with Python, Matplotlib, identifying a 12% average monthly churn rate and 17% new seller adoption rate.

Engineered prompt library for AI-driven content creation; automated blog generation from interview transcripts using ChatGPT and Claude, reducing task time by 70% and achieving 80% quality match to human-written content. • Developed comprehensive AI chatbot comparison matrix; analyzed 20+ solutions across 10 key factors, enabling data-driven selection of optimal marketing assistance tools and reducing decision-making time. • Conducted cross-departmental interviews to identify Generative AI opportunities; proposed tailored solutions to enhance marketing efficiency demonstrating strong analytical and problem-solving skills. • Spearheaded AI literacy initiative, with presentations, workshops for 60+ employees; increased understanding of Generative AI and LLMs by 75%, enabling informed decision-making on AI integration projects.

Implemented a Q-Learning based reinforcement learning method for Analysing frictions in Generalised Second Price Auctions & sellers’ bidding behaviours • Studied cutting-edge techniques such as Heterogeneous Treatment Effects, Causal Tress and Random Forests, alongside machine learning methodologies, to analyse & unravel complex patterns in historical auction data • Conducted an extensive literature review to understand bidding behavior, auction mechanisms, budget constraints, and real-time auctions in the context of sponsored search advertising

Optimized the software DeepFaceLab which is used create DeepFakes using OpenCV , Google Colab • Created a fast and efficient version of the software which can work without use of Graphical Processing (GPU) • Integrated the original code of Image Face extraction [S3FD] with a Caffe Model and improved the algorithm for face landmarks detection & face segmentation, reducing the running time by 80% compared to original time . • Optimized the training [SAEHD] and face alignment pipeline of the software, providing a better DeepFake result as well as reducing the time of complete execution by more than 65%