Rudresh Mehta

Architected an AI-powered product-search system based on RAG using Azure OpenAI and Pinecone, processing 100K+ product queries daily. Delivered natural language search capabilities that increased product discovery and generated over $3M in additional revenue through semantic search with hybrid dense/sparse vector retrieval. Automated Windows server provisioning via NLP-powered AI agents and self-hosted LLMs, saving substantial manual effort and costs. Built production ETL pipelines across AWS Redshift and Azure Synapse with 5TB+ daily transactions and 99.8% success rate; reduced ETL processing time by 30% through parallel processing and optimized transformations. Implemented data quality frameworks with DBT ensuring data integrity and preventing downstream errors worth $1.2M. Led cloud migration of 15TB of legacy databases to Azure Synapse and Microsoft Fabric. Developed feature engineering pipelines on Microsoft Fabric (PySpark) handling 10M+ daily transactions, accelerating model dev

Architected an AI-powered, product-expert solution based on Retrieval-Augmented Generation (RAG) leveraging Azure OpenAI and Pinecone. It processed 100K+ product queries daily, delivering natural language search capabilities that enhanced product discovery and generated over $3M in incremental revenue through semantic search with hybrid dense and sparse retrieval. Automated JIRA-based Windows server access provisioning using NLP-driven AI agents and self-hosted LLM/GPT-OSS on internal infrastructure, reducing manual effort. Built end-to-end production ETL pipelines with Apache Airflow orchestrating data workflows across AWS Redshift and Azure Synapse, processing 5TB+ daily financial transactions with 99.8% reliability, cutting ETL time by ~30% (8h to 5.5h). Designed data quality framework with DBT including line-level tests, checksum validation, and schema compatibility checks that reduced data integrity issues by 95% and prevented roughly $1.2M in downstream errors during migrations.

Developed real-time fraud detection systems using ensemble ML models processing 5M+ transactions daily with low latency; engineered features on transaction patterns, geo-location anomalies, and behavioral signals, achieving high detection accuracy while preventing losses exceeding $26M annually with minimal false positives. Built a credit risk scoring engine using gradient boosting and deep learning on a decade of historical loan data with hundreds of engineered features, improving default prediction. Architected a customer churn prediction platform using NLP sentiment analysis on support interactions and behavioral analytics, deploying optimized LSTM models on Azure ML to identify at-risk customers early and enable proactive retention campaigns saving over $3M annually. Implemented automated KYC validation combining CV for document verification and NLP for entity extraction to handle thousands of onboarding requests with improved straight-through processing.