Tushar Chauhan

Collaborated with a 4-member team to design and implement an AI-driven product recommendation system, aiming to reduce manual effort and enhance user experience. Analyzed sales patterns, identified top-selling products, and performed data preprocessing. Engineered features, incorporating sales, quantity, frequency, recency, seasonality, geography, and promotion aspects for each retailer’s products. Developed a Random Forest multi-output classifier to predict the probability of being sold in the market. Implemented sales pattern-based strategies and business rules tailored to meet market requirements.

Oversaw the crowd control and security issues that arose during the festival. Assisted the Core Team in disbursing its duties and provided the input to the members of the festival core team in all tiers for resolution of conflicts.

Managed a team of three members to develop scorecards for assessing default risk of new customers applying for home loans and credit cards. Conducted feature selection using XGBoost’s variable importance and correlation matrix. Generated feature bins to calculate the Information Value and Weight of Evidence (WoE) for each variable. Implemented a generalized linear model (GLM) in R to capture the defaulted customers. Analyzed the outcomes by creating actual and predicted trends. Led and collaborated with a 2-member team to develop a computer vision architecture for extracting iron ores from high-resolution satellite imagery. Gained comprehension of fundamental concepts of TFRecords and utilized them to optimize computations. Conducted research and implemented various image segmentation techniques to achieve desired results. Utilized transfer learning from pre-trained models like RESNET, DENSENET, VGG, and MOBILENET. Implemented the U-NET architecture to perform accurate segmentation of iron-ores in the satellite imagery. Developed a credit risk detection model and capture diverse purchasing patterns of various customer segments. Created XGBoost and Neural Network models using customer’s historical transaction data and credit history. Engineered features at both customer level and transaction level to enhance good/bad capturing model capacity. Worked with databrick clusters and azure databricks environment. Fine tuned the model’s hyper-parameters using hyper-optimisation.