Infrastructure I run, and products I've shipped.

Platform and reliability work across healthcare, financial data and telecom, plus live consumer products where I own the whole stack — from Terraform to the front end.

Infrastructure and platform work

Multi-cluster Kubernetes platform (Capsule)

Operate six production EKS clusters with Helm, Helmfile and Codefresh CI/CD, Kyverno security policies, and SLO/SLI-driven burn-rate alerting.

  • 99.9% uptime
  • Zero-downtime EKS 1.32 → 1.33 → 1.34 upgrades
  • ~20% faster deployments
  • Kubernetes
  • EKS
  • Helm
  • Codefresh
  • Kyverno
  • Prometheus

Aurora → RDS PostgreSQL migration

Led a cost-optimising migration from Aurora Serverless v2 to RDS Provisioned, plus a zero-downtime RDS encryption migration to customer-managed KMS keys.

  • ~8% database cost reduction
  • Zero downtime
  • Customer-managed encryption
  • AWS RDS
  • PostgreSQL
  • KMS
  • Terraform

Reusable Terraform and security pipeline

Authored a reusable Terraform module for Temporal workflow orchestration adopted by 10+ teams, and a shared security pipeline scanning every repo.

  • Adopted by 10+ teams
  • Org-wide security scanning
  • Automated branch and ECR policies
  • Terraform
  • Checkov
  • Hadolint
  • Snyk
  • Codefresh

1TB MySQL → AWS RDS migration (FactSet)

Migrated a 1TB MySQL MyISAM database from on-prem to AWS RDS using rsync and AWS DMS, with a checkpoint-based resumable ingestion pipeline.

  • 1TB migrated
  • Checkpoint-based resume
  • ~10% AWS cost reduction
  • AWS DMS
  • MySQL
  • InnoDB
  • Python
  • Bash

Telecom network automation microservice

Built a Python/Flask hardware-abstraction microservice with SR-IOV and connectivity APIs, plus a self-update API across 100+ telecom nodes.

  • 100+ nodes managed
  • Remote fleet self-update
  • Metrics collection APIs
  • Python
  • Flask
  • KVM
  • Linux
  • systemd

Observability and SLO framework

Designed an SLO/SLI framework with automated burn-rate alerting and long-term metrics retention across clusters using Prometheus, Grafana, and Thanos.

  • Automated burn-rate alerts
  • Long-term metrics retention
  • Multi-cluster observability
  • Prometheus
  • Grafana
  • Thanos
  • Alertmanager

Products and client work

All live. Each card opens the site.

CardKhata

Live

Forward an Indian credit card statement by email and get every transaction categorised, every EMI rebuilt, every fee itemised, and annual fee-waiver progress tracked. Deterministic issuer parsers run first and a language model is the fallback only when no parser matches — roughly a 30x cost difference against an LLM-first design. Statements arrive through SES inbound, land SSE-KMS encrypted in S3, and are parsed by a container Lambda.

  • Email-in statement ingestion via SES
  • EMI reconstruction and fee-waiver tracking
  • Deterministic parsers first, model fallback second
  • Next.js
  • TypeScript
  • PHP
  • Python
  • AWS Lambda
  • SES
  • S3
  • KMS
  • RDS MariaDB
  • Route 53
  • Terraform
cardkhata.in