Rishi Vamshi Athinarap

About Rishi Vamshi Athinarap

I build backends that hold up under ugly, real‑world scale. At Illumio CloudSecure I own parts of Cloud Inventory, the multi‑cloud inventory/graph engine behind the product. I care about preventing issues before they page us: I designed a distributed, hierarchical client‑side rate limiter that turned months‑long initial syncs into ~1‑day updates for very large tenants (tens of millions of resources), and rewrote our metadata/storage path to cut per‑tenant memory from ~4–10 GB to <100 MB.I also like making teams faster. I built an AI‑assisted “Cloud Sync Agent” (embeddings + Go AST + a small MCP server) that safely scaffolds new resource‑type integrations straight from cloud SDKs, reducing week‑long work to hours while keeping guardrails tight.Before Illumio, at MRI Software I helped ship the first production AI QA engine in PropTech, for legal/real‑estate documents (Summer ’23). We scaled to hundreds of thousands of pages, added smarter chunking and correctness checks, and delivered sub‑second answers via vector + hybrid search on Elasticsearch.Earlier, I completed an MS in CS (UNC Charlotte) and worked as an RA with Prof. Wang on real‑time kinematic sequence analysis for athletes.Toolbox: Go · Postgres (RDS/Azure) · Neo4j · Kafka · Redis · AWS/Azure/GCP/OCI · Elasticsearch · Vector/hybrid search · LLMs/embeddings · AST tooling.

Experience

1. Senior Member of Technical Staff

   Illumio

   Aug 2025 — Present · Sunnyvale, CA, US

   Working on Scaling CloudSecure, multi‑cloud asset & relationship inventory.I’m a distributed backend engineer scaling CloudSecure’s multi‑cloud inventory/graph—and I use AI where it removes toil and speeds delivery.Made “months → ~1 day” possible: designed a distributed, hierarchical client‑side rate limiter with custom limits and scoring algorithm for tenant, account, and other pluggable parameters. Super light weight, added gRPC interceptors so calls stay just under AWS/Azure/GCP quotas, scaling to sync 40M+ resources in record time. AI that removes toil: built the Cloud Sync Agent (embeddings + Go AST + a small MCP server) that scaffolds new resource‑type integrations directly from cloud SDKs, cutting “week‑long” work to hours with guardrails that keep changes safe.Cut memory ~97–99%(≈4–10 GB → <100 MB per tenant) by rewriting metadata/storage: diff‑based updates, coroutine fan‑out, and paginated/batched Postgres writes.Compliance (Novel Agent Pipeline): prototype that parses/dedupes CIS/NIST/AWS plus 40+ benchmarks and frameworks, maps rules ↔ inventory, and autogenerates checks. Won an internal hackathon due to breakthroughs of implementing Agent workflow at scale. Postgres at write‑heavy scale: custom autovacuum policies, pg_stat* analysis, index redesign, multi parameter tuning on RDS/Azure Postgres, and tenant‑aware partitioning → stable p95s and no OOMs under ingest.Faster graph updates: decomposed work to add Resource Type into each task and began moving from Redis to Kafka for near‑real‑time relationship updates in Neo4j.Reliability & observability: standardized structured logging with a shared field schema, trimmed high‑cardinality metrics, and added dashboards — our area consistently kept Sev‑2s low across the entire product. GCP launch: shipped ~25 resource types in a week;~55 at GA. Re‑architected for global resources while keeping graph integrity and honoring data residency.

Education

• University of North Carolina at Charlotte

  Masters of Science, Computer Science

• Vellore Institute of Technology

  Bachelor of Technology - BTech, Computer Science

  2015 — 2019