Sadhana Geddam Edison

Build an agentic triage and routing service for the HPE GreenLake cloud platform using the Anthropic API, LangGraph, and MCP servers, improving mean resolution time across high daily interaction volume. Implement RAG over internal docs/runbooks with pgvector and Azure Cognitive Search, tuning chunking and reranking to raise retrieval precision. Gate prompt/model/retrieval changes behind LangSmith evaluation datasets and regression checks in CI/CD to ensure measured quality and 99.9% uptime. Develop Java/Spring Boot and Go microservices behind REST and GraphQL APIs, moving bursty work to Kafka with idempotent consumers and DLQs to reduce p95 latency.

Took a LangGraph multi-agent platform from prototype to production for 500+ active users, owning requirements, design, deployment, monitoring, and support. Built MCP servers with FastMCP to expose university systems as tools/resources/prompts and ran agents on Amazon Bedrock AgentCore with human-in-the-loop approval for risky actions. Created an LLM evaluation harness with automated judging to score similarity and drive retrieval/prompt changes before release. Built FastAPI and Node.js/TypeScript backends with rate limiting, token cost tracking, Redis caching, and provider routing; improved API latency and ensured accessibility compliance. Added merge gating with GitHub Actions tests and included PII redaction to prevent personal data from reaching models or leaving approved systems.

Automated 100+ REST API health checks across a cloud validation platform and integrated them into CI/CD as a release gate to reduce production deployment errors. Built cloud-native services on AWS (EC2, EKS, Lambda, API Gateway, RDS, S3) in Python and Go, maintaining 99.9%+ uptime across regions and improving incident response time.