Divy Patel

Built a Retrieval-Augmented Generation (RAG) solution using LangChain, OpenAI APIs, and a vector database, improving internal knowledge retrieval accuracy by 11% during pilot deployment. Engineered automated ML pipelines with MLflow and Docker to reduce model retraining cycle time by 9% and improve reproducibility across teams. Deployed a FastAPI-based inference service on AWS, lowering average prediction latency by 8%. Optimized feature engineering and hyperparameter tuning, increasing pilot model F1 scores by 7% on classification tasks. Created explainable AI reporting and Power BI dashboards to help stakeholders identify key drivers and shorten model review cycles by 12%.

Led development of a Retrieval-Augmented Generation (RAG) solution using LangChain, OpenAI APIs, and a vector database, improving internal knowledge retrieval accuracy by 11% during a pilot. Built automated ML pipelines with MLflow and Docker to reduce model retraining cycle time by 9% and improve reproducibility across teams. Orchestrated a FastAPI inference service on AWS, reducing average prediction latency by 8% for a customer-facing analytics application. Improved feature engineering and hyperparameter tuning workflows, increasing pilot classification F1 scores by 7%. Created explainable AI reporting and Power BI dashboards, helping stakeholders identify key drivers and shortening model review cycles by 12%.

Developed predictive machine learning solutions with Python, scikit-learn, and SQL, improving classification accuracy by 19% across multiple client projects. Built ensemble models for demand forecasting, improving forecast accuracy by 24% versus legacy methods. Designed ETL and feature pipelines with Python and Snowflake, reducing data preparation effort by 28%. Deployed deep learning and time-series solutions on Azure, cutting manual planning time by 18%. Integrated Hugging Face Transformers for text analytics, improving document classification precision by 20%. Implemented CI/CD practices using Git and Docker, shortening model release cycles by 22%. Used statistical testing and A/B experimentation to validate model improvements, raising engagement metrics by 14%. Mentored junior analysts on MLOps and model evaluation, improving first-pass deployment success by 16%.