Harshal Mahakalkar

Managed advertising budgets of EUR 1M+ across multiple retail and fashion clients, translating campaign and revenue data into budget allocation and growth decisions. Built SQL datasets and automated data cleaning, reconciliation, KPI calculations, and anomaly checks using Python-based data processing. Created recurring dashboards and validated multi-channel tracking across ad platforms to improve reporting reliability, and applied forecasting plus clustering approaches for scenario planning and segmentation.

Delivered end-to-end machine learning workflows for structured and text data, including preprocessing, EDA, feature engineering, and model training. Implemented classification, prediction, clustering, and sentiment analysis with train-test validation and metric-based comparison of results. Interpreted model outputs and supported iteration based on evaluation findings.

Implemented a TensorFlow/Keras image classification pipeline for categorizing clothing. Covered labeled data ingestion, preprocessing, neural network training, and evaluation with loss/accuracy monitoring. Added prediction and visualization steps to review model performance.

Built a desktop AI support system for dementia patients using face recognition and vision pipelines. Implemented LBPH and Haar cascade based recognition with OpenCV and Dlib, and added NLTK-based text processing. Used an SQL-backed data layer to retrieve contacts and reminders through a Tkinter interface.

Developed a customer segmentation workflow using clustering on a large customer dataset. Performed exploratory analysis and feature-level preparation, then trained a K-means clustering model to derive behavioral segments. Visualized segment characteristics to support targeted marketing and customer profiling.