Experience

1. AI Data Engineer (FinTech)

   EXL Analytics

   March 2025 – Present · Mumbai, India

   

   • Designed a Multi-Agent RCA Platform on Databricks using LangGraph, LangChain, Claude, and Milvus, with natural-language-to-SQL over Unity Catalog reconciliation datasets — cut root-cause analysis from hours to under 5 minutes.
   • Built production RAG pipelines with LangChain and open-source LLMs over 1 TB of financial data, improving retrieval speed by 30% and reducing latency by 25%.
   • Engineered a hybrid retrieval framework (RAG + MCP) combining graph and keyword search, increasing operational responsiveness by 40% through faster query resolution.
   • Implemented SBERT-based semantic ranking, reducing memory usage by 28% and lifting successful query resolution by 18%.
   • Deployed Llama 3.3 70B in production on vLLM with continuous batching, serving 128 concurrent requests across 10+ applications.

2. Data Engineer (Reliance Intelligence)

   Jio Platforms Limited

   December 2023 – March 2025 · Mumbai, India

   

   • Architected the high-throughput real-time ingestion pipeline for Jio AirFiber with Kafka, Spark, Iceberg, and Delta Lake, processing 1.2 billion records per hour — cut troubleshooting response time by 80%.
   • Built real-time monitoring and alerting with Prometheus and Grafana for Kafka consumer lag, Spark job health, and pipeline throughput — cut mean-time-to-detection of failures by 90% and enabled resolution before SLA breaches.
   • Designed a metadata-driven data platform on Databricks and Azure for Jio Financial Services, standardizing ingestion and transformation across 7+ data domains — cut ad-hoc pipeline development time by 90%.
   • Re-engineered 9 streaming KPIs for Reliance Life Sciences: replaced Redis with HDFS (memory −30%, performance +25%) and optimized Spark window functions to cut shuffling by 40%.
   • Led the Synapse-to-on-premise (S2O) migration of 28 streaming jobs and 13 KPI Spark jobs under a unified driver — 25% faster processing, 40% higher throughput — with Debezium CDC feeding real-time analytics on Apache Druid.