Pravash Purkay

Data Engineer

Pine Labs • Temps plein

Jul 2023 - Jul 2026 • 3 yrs

Built and deployed a production-grade NLP system using RAG, LLMs, and semantic search to auto-generate validated SQL queries from natural language inputs across 1,800+ enterprise tables — applying embedding-based retrieval, contextual ranking, and transformer-based text understanding at scale. – Designed end-to-end ML pipelines for enterprise analytics automation using Python, LangChain, LangGraph, and ChromaDB, reducing average query resolution time from 5–7 days to under 30 seconds — a 99%+ latency improvement across 50+ business users. – Applied statistical modelling and anomaly detection techniques to merchant transaction data across 10+ KPIs— identifying revenue trend deviations, refund rate anomalies, and settlement pattern irregularities using SQL window functions and Python-based statistical tests. – Implemented feature engineering and data preprocessing workflows over 56,000+ columns of enterprise transaction data — including missing value imputation, schema normalisation, and embedding-based dimensionality reduction for downstream ML and retrieval tasks. – Built AI-powered self-serve analytics and predictive reporting tools using Streamlit, Plotly, and AWS S3, reducing decision-making turnaround from 1 week to 1 day (60% improvement) and improving ticket resolution efficiency by 95%. – Designed and maintained interactive BI dashboards in Apache Superset tracking merchant performance — used weekly by 5+ cross-functional teams for data-driven strategic decisions. – Led migration of 100+ data tables to Amazon Redshift cloud architecture, optimising query performance by 40% and reducing annual infrastructure costs by approximately $61,560. – Developed model evaluation and validation frameworks for LLM-generated outputs — including benchmark design, accuracy scoring, and failure mode analysis — achieving 87%+ first-attempt accuracy across a 60-question test suite.