vedavathi thumula

About

Backend/platform engineer who has owned a Python/FastAPI results API and deployed it on Kubernetes with Helm and GitHub Actions-driven CI/CD. Demonstrates strong production operations mindset across performance tuning, monitoring, safe rollouts/rollbacks, and phased migrations, plus hands-on Kafka streaming experience focused on ordering and idempotency.

Experience

Key Strengths

• Owned end-to-end design and build of a FastAPI results API backed by PostgreSQL/SQLAlchemy
• Performance and reliability improvements via query optimization, caching, and background tasking
• Production-grade Kubernetes deployments using Helm with health probes, rolling updates, and fast rollback
• Structured migration execution (phased plan, parallel testing, incremental rollout, monitoring/rollback)
• Kafka streaming reliability patterns: keyed partitions for ordering, idempotent consumers/writes, offset discipline
• GitOps implementation with infra-as-code in Git, PR-based change control, and automated deployments via GitHub Actions
• Takes LLM prototypes to production with clear success metrics (accuracy, latency, cost) and SLA focus
• Builds reliability via prompt hardening, guardrails, RAG with trusted sources, and automated evals to prevent regressions
• Strong production troubleshooting: isolates data vs prompting vs model vs orchestration using logs and live metrics
• Rapid incident response on retrieval failures (tightened filters, relevance thresholds, fallbacks) to stabilize quality
• Measures post-deploy success with offline evals plus live monitoring and user feedback signals
• Effective developer enablement through hands-on demos/workshops focused on architecture, APIs, and failure modes
• Partners with sales on discovery, tailored demos/POCs, and real-time objection handling to accelerate adoption
• Designed and deployed production-grade GenAI RAG + agentic backend (FastAPI microservices, FAISS, LangGraph) on Azure Kubernetes
• Latency optimization for high-volume GenAI APIs (async endpoints, caching, optimized embeddings)
• Scalable API operations (Docker/Kubernetes, autoscaling, load testing, CI/CD)
• Secure multi-tenant architecture (JWT/OAuth, RBAC, database row-level security, audit logging)
• Zero-downtime refactors/migrations from monolith to microservices using versioning, contract tests, parallel runs, staged traffic shifts, and rollback
• Improved agent reliability by identifying tool-response edge cases and adding validation, retries, fallbacks, and structured error handling
• Built and deployed production LLM agent system automating multi-step internal workflow (Walmart)
• Reduced agent latency by >50% via distillation, ONNX optimization, caching, and DAG parallelization
• Improved pipeline reliability using DAG control flows, tool validation, retries, and deterministic fallbacks
• Orchestrated production AI workflows with Airflow/Prefect/LangGraph; cut processing time >40%
• Metrics-driven iteration (latency, success rate, grounding accuracy, failure-mode tracking) leading to >33% latency reduction and higher success rate
• Modular system design enabling model/retrieval/prompt strategy swaps as requirements evolve
• Effective collaboration with non-technical stakeholders using demos/mockups and clear trade-off communication