Indra K Dutta

Built production agentic AI workflows in Microsoft Semantic Kernel with handoff and group-chat orchestration across specialized LLM agents. Integrated Azure AI Search as the retrieval layer grounded in enterprise financial data, improving search response times and supporting high query volume. Architected a new AWS landing zone, leading migration of 50+ workloads off on-prem and legacy zones with cost reduction. Led development of microservices and data ingestion following Medallion-style architecture, including Python/PySpark transformations and AWS Glue ETL. Deployed containerized services to Amazon EKS with automated CI/CD and instrumented observability using Dynatrace.

Engineered an end-to-end RAG pipeline in Python using LangChain and Azure OpenAI to improve processing efficiency while handling very large record volumes with high accuracy. Enhanced retrieval using metadata-enriched embeddings and filtered vector search. Reduced LLM latency and inference cost via semantic caching, batched chunk writes, and model tiering. Modernized a commodity trading workflow into a cloud-native platform, and implemented large-scale Medallion-style data processing on Snowflake with dbt. Communicated architecture trade-offs and delivery progress to stakeholders.

Implemented natural-language RAG search over a SQL database in Python using LangChain and OpenAI, enabling conversational querying of betting data. Applied prompt engineering patterns (CoT and ReAct) to improve reasoning and tool-use reliability. Built conversational AI experiences with Azure chatbot services. Developed microservices for streaming real-time sports data into the Caesars system at high transaction rates, including AWS Lambda and Kafka delivery. Built a React.js customer-facing front end with real-time odds and markets using PrimeReact.