AI/ML Engineer at Kroger

Deployed a calibrated LightGBM cart-return propensity model with a nightly batch scorer to the data lake and customer-level scores; added AUC and drift monitors; used scikit-learn for preprocessing and isotonic calibration. Enhanced customer segmentation with supervised propensity modeling in TensorFlow and Keras and unsupervised k-prototypes and HDBSCAN for targeted promotions. Augmented the recommendation system by combining matrix factorization for candidate generation and item-item collaborative filtering; added cold-start coverage with Hugging Face sentence-transformer embeddings in PyTorch; improved Hit@K and NDCG offline and supported A/B evaluation under predefined guardrails. Applied NLP with NLTK and Hugging Face Transformers to extract aspect-level sentiment from product reviews and integrated a review-based score into search ranking and early inventory alerts. Piloted a Gen AI and NLP policy assistant using an LLM with RAG over policy and FAQ documents with embeddings and a

AI/ML Engineer at Kroger

Deployed a calibrated LightGBM cart-return propensity model with nightly batch scoring to the data lake and customer-level scores, with AUC and drift monitoring. Implemented supervised propensity modeling in TensorFlow/Keras and unsupervised clustering (K-prototypes, HDBSCAN) to refine customer segmentation and targeted promotions. Enhanced the recommender system with matrix factorization, item-item collaborative filtering, and cold-start coverage using sentence-transformer embeddings for offline evaluation and A/B guardrails. Applied NLP on reviews to extract sentiment and integrated a review-based score into search and inventory alerts. Piloted a Gen AI/NLP policy assistant using an LLM with RAG and FAISS, including PII redaction and citations. Added graph-derived features to improve fraud scoring with CatBoost and implemented time-aware cross-validation with SHAP explanations and KS/PR-AUC monitors.