Aman Kumar

• Created a complex hierarchical forecasting model pipeline using Kedro for in-season forecasting with very limited data • Enacted a reconciliation strategy for coherent forecasts at three different hierarchical levels respectively • Done hyperparameter tuning using Optuna, integrated its dashboard and MLflow for model tracking and management • Optimized production quantities and minimized excess inventory by providing a Cut Plan for denim jeans leveraging forecasting model using sales data and market trends • Developed a poc of footfall analysis using CCTV footage, YOLOv8 for object detection, and object tracking • Evolved a data visualization and exploration tool using Apache Superset • Generated multiple interactive dashboards using PowerBI and Superset, enhancing data visualization and decision-making

• Utilized OpenAI’s Whisper model to transcribe conversations between two people into text with high accuracy • Integrated the transcribed text into the Salesken platform to enhance real-time analytics and insights • Employed Hugging Face’s T5 model to summarize transcriptions and extract useful insights from conversations • Conducted thorough testing and validation of transcription and summarization models to ensure high performance and accuracy • Enhanced the platform’s ability to provide actionable insights from conversation data, improving decision-making processes

• Selected in the integrated learning course with practical industry applications through an immersive experience with Amazon Scientist • Inculcated knowledge about some key ML topics including Supervised Learning, Deep Neural Networks, Dimensionality Reduction, Unsupervised Learning, Probabilistic Graphical Models, and Sequential Learning

Supervisor - Prof. Manoj Kumar Tiwari, Director IIM MUMBAI (previously NITIE Mumbai ) • Implemented a mathematical model GDMR from scratch to capture the purchase trends in sale forecasting • Predicted price of products using numeric, textual and 12 weeks google trends data using various machine learning models • Reduced 12% error, by integrating the GDMR with Lasso, XGBoost and SVM Regression through stacking ensemble model

• Created a Credit Profile of the user with multiple features using an ensemble model for sanctioning the loan amount to bearer • Performed Semi-supervised learning on SMS dataset using Snorkel and resampled the data to avoid imbalance in dataset • Achieved an F1-Score of 0.94 on ensemble model of Multinomial Naive Bayes, K-Nearest Neighbors and Random Forest