Jagadish Rapuru

Built an agentic invoice processing pipeline in .NET 8 and Blazor Server that reads subrecipient invoices with Azure Document Intelligence and routes approvals to finance staff, saving 18 hours per month across 600+ invoices. Designed a multi-agent workflow with Semantic Kernel and Azure OpenAI GPT (extraction, coding, approval routing, document generation) coordinated by a C# state machine. Added backup validation using Azure AI Foundry vision models with structured output schema to compare receipts/bank statements against invoice totals and route low-confidence cases to human reviewers instead of auto-approving. Implemented an agentic RAG assistant in Python (LangChain/LangGraph/ChromaDB) to answer questions with citations and an escalate loop. Integrated with Azure Logic Apps, Microsoft Teams, SharePoint, and a Workday EIB feed so invoices arrive, are reviewed, and post to the ERP without email back-and-forth. Implemented agent audit logging and a golden-label test set to measure fi

Built the GBCESU research funding portal in Blazor Server with Radzen UI to track federal awards, subaward budgets, and expenditures across 10+ institutions and 150+ funded projects. Automated processing of award notices using Azure OpenAI GPT and Semantic Kernel agents to extract funding amounts and budget periods and flag mismatches for finance review. Added JWT authentication and role-based authorization so faculty, finance administrators, and external collaborators see only permitted data. Wrote SQL Server views and SSIS packages to consolidate award and expenditure data from five university systems into a reporting database, replacing manual spreadsheet preparation for monthly reports. Delivered Power BI dashboards with anomaly detection using scikit-learn Isolation Forest to flag unusual monthly spend. Implemented Carnegie Classification expenditure reporting views using optimized SQL for yearly research assessment and accreditation.

Developed REST APIs in ASP.NET Core with Angular front ends for an energy asset management platform used by utility clients to track substations, transformers, meters, and field equipment, owning 12+ backend services. Migrated legacy asset and maintenance records from Excel into the platform using .NET console apps, Python (pandas), and SQL scripts with data cleanup and feature engineering feeding a scikit-learn load forecasting model. Trained a CNN classifier in PyTorch to flag faulty and at-risk substation equipment and provide a ranked list to plan maintenance work orders. Implemented PL/pgSQL functions and stored procedures for multi-filter search over 100K+ equipment, meter reading, and outage records to support grid operations reporting. Added Redis caching for reference data used frequently by Angular dashboards to cut redundant database calls.