At Bisviews, I sourced and verified 7M+ business listings, then designed a transformation pipeline and semantic classification system to structure the records. I also architected a PostgreSQL dimensional warehouse with sub-second query performance.
As an independent Data Scientist and ML Engineer, I fine-tuned a Vision Transformer across 38 PlantVillage classes for crop-disease diagnosis. Grad-CAM validation helped ensure predictions focused on lesion patterns; I deployed the model through FastAPI, Gradio, and Hugging Face Spaces, reaching 94.3% accuracy with inference under 200ms.
For financial fraud detection, I designed a real-time pipeline using Kafka and XGBoost that processed 500K+ daily transactions at sub-100ms latency, with SHAP explanations and 99.1% precision. I’ve also built semantic document search with GPT-4o, LangChain, and Pinecone, and a time-series ensemble that reduced MAE by 34% against an ARIMA baseline.
My personal research explores responsible ML and bias mitigation for low-resource agricultural datasets, alongside a climate-smart AI assistant using Llama 3 8B and real-time agricultural data. I also research diffusion models for augmenting imperfect remote-sensing imagery.

