Conducted sentiment analysis on product reviews for a movie streaming service. Preprocessed and vectorized non-traditional data using techniques such as Bag of Words (BoW), TF-IDF, and word embedding model. Developed a sentiment analysis model to identify positive and negative opinions in movie reviews. Conducted home credit risk analysis, predicting repayment capability for home credit applicants. Manipulated, visualized, and pre-processed a dataset of over 350,000 transactions. Trained supervised models (DecisionTree, XGBoost, LightGBM) achieving ROC AUC over 0.72. Analyzed revenue and delivery metrics for a leading Latam E-commerce site during 2016-2018. Identified revenue trends, popular product categories, and state-wise revenue distribution. Assessed delivery performance by analyzing package delivery time and estimated vs. actual delivery dates. Developed a data pipeline (ELT) using SQLite and SQL, utilized Matplotlib and Seaborn for visualizations, and generated actionable insights into revenue patterns, product popularity, and delivery performance. Implemented image classification for e-commerce, predicting vehicle make and model from unstructured images. Cleaned and visualized the dataset, performed data augmentation, and trained a fine-grained classification model using convolutional neural networks (CNN). Achieved 82% accuracy and deployed the model as an API-based web-service application on AWS instances using Docker.

Achieved 95% accuracy with GCNN for action recognition in clinical settings, utilizing Python, OpenCV, PyTorch, and SciPy. Applied the GCNN in clinical settings to monitor patients and guide medical decisions while ensuring patients' privacy. Improved hyperspectral image retrieval by +2dB in Peak-to-Signal Noise Ratio through coded aperture architecture optimization, using PyTorch, Numpy, and Matplotlib. Enhanced seismic data reconstruction with CNNs, yielding a remarkable 12dB gain in Peak-to-Signal Noise Ratio and a 0.5 increase in Similarity Index. Reduced data acquisition costs by 30% via effective data estimation, employing TensorFlow and Numpy.