Sarvagya B.

About Sarvagya B.

I’m a Software Engineer specializing in backend infrastructure, cloud-native systems, and applied AI in observability, with hands-on experience in multi-tenant, production-grade deployments.At Cisco, as a Forward Deployed Engineer, I engineered a multi-agent LLM orchestration platform for observability workflows, integrating retrieval-augmented generation (RAG) pipelines combining hybrid semantic and keyword search. I implemented real-time WebSocket services, Thrift-based RPC pipelines, and GraphQL/REST integrations, while optimizing multi-region Kubernetes infrastructure to sustain sub-300ms P95 latency globally. My work spans Kubernetes telemetry, RCA pipelines, and real-time feature flag systems, combining cutting-edge AI with enterprise-grade reliability, security, and multi-tenant isolation.Previously at UnitedHealth Group, I modernized monolithic applications into Spring Boot microservices on Azure Kubernetes Service, built Kafka-based real-time streaming pipelines, migrated databases to Azure CosmosDB, and streamlined CI/CD with Terraform and Jenkins. I also mentored engineers on cloud-native and distributed workflows.During my research in distributed ML and edge computing, I scaled PyTorch training workloads across Kubernetes clusters, built distributed deployment pipelines in Golang/Docker, and developed Prometheus/Grafana/Loki dashboards for cluster observability, improving reliability and monitoring efficiency.I thrive at the intersection of backend infrastructure, cloud-native engineering, and intelligent automation, building systems that scale reliably, simplify operations, and make troubleshooting smarter. I’m passionate about combining observability, AI/LLM experimentation, and production-grade engineering to solve real-world enterprise problems.

Experience

1. Forward Deployed Engineer

   Cisco

   Jun 2024 — Present · NY, US

   Engineered a multi-agent LLM orchestration framework with domain-specific sub-LLMs (infrastructure, logs, metrics) to support observability and incident diagnostics, cutting investigation time by 35%.Spearheaded multi-region infrastructure expansion, optimizing resources and leveraging regional latency insights to maintain consistent 300ms P95 response times globally.Integrated OpenAI streaming APIs with custom adapters supporting reasoning and standard models, embedding telemetry and robust error handling for enterprise-grade reliability.Built RAG pipelines with vectorized retrieval over telemetry and documentation embeddings, enabling semantic search and surfacing relevant charts/metrics reducing incident triage time by 30%.Implemented RBAC-aware dynamic LLM prompts with step-by-step reasoning, ensuring users only access permitted tools and receive reliable, actionable insights.Spearheaded DeepEval integration within LLM workflows to detect low-confidence or hallucinated responses, ensuring high-quality, reliable insights delivered to engineers and automation systems.Developed Kubernetes telemetry and RCA pipelines, combining incident timelines and chat-driven diagnostics to streamline cross-service root cause discovery.Engineered a real-time feature flag system enabling safe rollouts, rapid experimentation, and on-call updates, minimizing downtime without full deployments.

Education

• Ch. Baldev Singh Model School

  Class 12

  2015 — 2016

• San José State University

  Master's degree, Computer Science

• Bal Bhavan Public School

  Class 10

  2013 — 2014

• Delhi College of Engineering

  Bachelor of Technology (B.Tech.), Computer Software Engineering

  2016 — 2020