CI/CD, Secrets and Access: GitHub Actions, self-hosted ARC runners, trunk-based release and rollback; Workload Identity Federation, GHE OIDC, IRSA + EKS Pod Identity, External Secrets Operator, cert-manager, Pomerium, Cloudflare Zero Trust, GKE Connect Gateway
Development & Data: Golang, Python, Bash, PostgreSQL, CloudNativePG, Aurora, ElastiCache, Starrocks, Redis, DMS, Pub/Sub, Linux
Professional Experience
SocureChennai, India
Senior Site Reliability Engineer (Remote)August 2026 - Present
Own the Fravity to Socure platform migration end to end: architected the target state, sequenced it across 5 teams, and cleared the account-boundary, PII and security sign-offs each phase depended on, with no release freeze for the teams shipping through it.
Built the GPU inference platform on EKS: Karpenter-scaled scale-from-zero L40S running vLLM on Ray Serve, Envoy AI Gateway plus Gateway API Inference Extension dispatching to llm-d, APISIX auth and rate limits, benchmarked to ~527K requests/day at FP8.
Made inference performance measurable: guidellm precision and traffic sweeps to p999 against chat-tier SLOs, GPUs halved by correcting an fp32 assumption to bf16, and the fine-tune tier sized on memory arithmetic proving 22 GiB a capability wall, not a preference.
Landed the whole platform on AWS across 104 merged PRs in 11 repos, replacing every GCP dependency with a managed equivalent: Cloud SQL to Aurora PostgreSQL, Pub/Sub to SQS, Workload Identity to IRSA; migrated 70 repos to GHE in < 5 days.
Took 4 services live end to end, then closed the defect class green deploys could not detect: config falling back to shipped defaults.
Shipped per-PR ephemeral environments for 4 components that need no cluster access and no SRE in the loop, scoped by an IRSA role granting inference only so unreviewed branches reach no storage, queue or secret; a 15-hour lockout class now self-heals in 5 minutes.
Own the observability contract for a platform with no agent and no ticket to add one: aligned it across platform SRE and two product teams, then shipped Datadog to 4 environments, ArgoCD in non-prod and Terraform in prod, 51 monitors, GPU tier on DCGM and vLLM latency.
Deployed the 3-environment EKS platform on /21 IPAM VPCs with TGW hub-and-spoke, then set the golden path: keyless GHE OIDC to ECR behind 4 scanners, Istio mTLS, FIPS golden bases, and 10 reusable Actions for trunk-based promotion with hotfix lanes.
One2NPune, Maharashtra, India
SRE Lead · Client: Fravity AI (Remote)May 2026 - July 2026
Built the org's first observability stack, then re-platformed it off AGPL onto VictoriaMetrics on a founder licence call: a dual-run Alloy fleet moved ~305K series from 192 targets in 4 clusters with no data-loss window, and metrics, logs, traces and alerts behind one Perses view.
Built fravity-spine, the internal Kubernetes platform other teams consume: private regional GKE, CMEK etcd, Workload Identity, CloudNativePG, OpenTofu across 6 state layers, and identity-based kubectl retiring bastions and shared SSH.
Led cost + access governance across 5 GCP + 5 AWS accounts as lead of a two-person SRE team: surfaced 19 live SA keys and ~$26,520/yr of list-price exposure, then shipped an org-wide FinOps board and IAM as Terraform on keyless CI.
Luxor TechnologySeattle, WA, US
DevOps Engineer (Remote)June 2023 - April 2026
Operated a multi-cluster Kubernetes platform for 20+ teams and 3,000+ enterprise customers: 8+ clusters across Hetzner, OVH, and GCP with 24/7 uptime for L1/L2 validator and RPC workloads, including 2 Hetzner bare-metal HA clusters (KubeOne, OpenEBS) built from scratch. Standardised on a 7-layer FluxCD GitOps pipeline with Linkerd mTLS, migrating 70+ secrets to Workload Identity.
Led incident response for a Litecoin mainnet hard fork: cross-team build, zero-downtime rolling deployment, postmortem. Restored production after a kubelet TLS expiry on bare metal, then shipped x509 monitoring across 10 clusters on 7-day and 24-hour PromQL alerts.
Owned CI/CD for 50+ repositories on self-hosted ARC runners. Ported a failing Playwright E2E suite from Mac hosted runners ($5,000/mo) to Linux ARC, cutting runtime 40 to 15 min and cost to $600/mo. Built code-signed Windows and Mac release pipelines on FluxCD.
Tata Consultancy ServicesHyderabad, Telangana, India
Systems Engineer (Remote)October 2020 - June 2023
Migrated 2,500+ repositories (GitLab, Bitbucket, SVN) to GitHub Enterprise for 120 teams and built their CI/CD pipelines, resolving self-hosted-API pagination limits across the platforms and saving 15 weeks of manual effort.
Built PowerShell, Python and Golang automation across 22 RHEL/WebLogic servers: monitoring, alerting, Oracle replication, and a storage-provisioning CLI that collapsed 30 man-hours of monthly tickets into one command, saving 380+ man-hours/month.
Open-Source Projects
ducat (June 2026 - Present) - Prometheus cost exporter in Python: cloud, SaaS + AI spend in one Grafana view (FOCUS spec).
Frit (May 2026 - Present) - GPU SRE lab: vLLM + Ray Serve on GPU Operator; inference-perf (TTFT/TPOT, DCGM) in Grafana.
Platform Zero (April 2026 - Present) - 4 SRE labs (k8slab, ansiblelab, awslab, gcplab); 0-to-100 in one command.
Truss (June 2026 - Present) - Self-hostable Supabase/Appwrite alternative, with an AI-agent MCP server.
florin (June 2026 - Present) - Portable DB workbench (VS Code): Postgres + Redis, connections + queries synced to a git vault you own.
Teaching Experience
Full-Stack Instructor, 4 cohorts (2022-2026), taught NextJS, FastAPI, and DevOps to BSc CS students.
Education
SASTRA Deemed UniversityMaster of Computer Applications (MCA) - Chennai, India - 2021 - 2023