At Ne•ia, I built and deployed sales-forecasting models on more than 1.2 million transactional records, reducing MAPE from 18.4% to 12.1%. I also shipped two models to production through reproducible ETL pipelines.
I segmented more than 210,000 customer profiles into three actionable cohorts used by marketing, and engineered features for fintech and healthcare use cases. I evaluated models with cross-validation and presented findings through Power BI and Matplotlib dashboards.
For my Intelligent Airbnb Search & Ranking Engine project, I built a two-stage retrieval and ranking pipeline using FAISS, Sentence Transformers, and XGBoost LambdaMART. It improved Recall@100 from 0.29 to 0.97 and NDCG@10 from 0.49 to 0.9997 on synthetic relevance labels.
I also developed AeroFuel AI, an explainable flight-fuel prediction and optimization project using EUROCONTROL data. My work included SHAP explanations, anomaly detection, and a FastAPI dashboard; I’m also a co-author of research on federated LLM personalization published at INSIGHT 2026.
