AI/ML Engineer at Databricks

Developed Python-based enterprise AI agents leveraging LangGraph, MLflow, Vector Search, and LLM orchestration serving 50,000+ business users across knowledge discovery, document intelligence, and workflow automation. Improved response accuracy by 32% using retrieval optimization, semantic chunking, embedding tuning, custom evaluation datasets, and automated feedback loops with MLflow evaluation and tracing. Reduced inference cost by 41% via dynamic model routing, prompt optimization, caching, and quality-cost tradeoff experiments across foundation model endpoints. Built multi-agent workflows in Python using LangGraph, LangChain, MCP, and tool-calling architectures. Designed and optimized RAG pipelines with Databricks Vector Search, Delta Lake, embeddings, and hybrid retrieval to improve contextual relevance and citation quality. Implemented MLflow-based evaluation frameworks (LLM-as-a-Judge), benchmarking, trace analysis, and hallucination detection for production reliability. Archite

AI/ML Engineer at Nvidia

Built Python-based network intelligence services using PyTorch, XGBoost, and time-series forecasting to predict congestion patterns across AI fabrics supporting 100K+ GPU-scale training environments. Implemented telemetry-driven anomaly detection using Python, Kafka, Spark, and Prometheus, improving detection accuracy by 28%. Designed adaptive routing optimization using reinforcement learning and graph analytics, reducing packet congestion latency by 35% and improving distributed training throughput. Implemented distributed ML workflows with PyTorch Distributed, NCCL, Ray, and Kubernetes to improve collective communication efficiency by 22%. Engineered real-time telemetry scoring and analytics microservices with FastAPI, gRPC, Redis, and PostgreSQL. Developed BlueField DPU offload services with NVIDIA DOCA, RDMA, and RoCE to reduce inference communication overhead by 30%. Built cloud-native observability on AWS (EKS) using Prometheus, Grafana, OpenTelemetry, and Kafka. Designed event-d