Certificate automation
adidas · cert-manager, Venafi, Helm, Kubernetes
100+ certs · zero missed renewals
THE PROBLEM
TLS, Kafka, ingress and domain certificates across dev, SIT, staging and production were requested and renewed by hand and tracked in spreadsheets. Ops had missed 10–20 renewals over time, usually found by users first.
WHAT I DID
Deployed cert-manager with Helm in every namespace and wired it to Venafi, so the security team's approval flow stayed in place. Teams now declare a certificate in Git; it's issued, stored as a Kubernetes secret and renewed before expiry with no one touching it.
Certificate in Git→cert-manager
per namespace→requestVenafi
approval→signedKubernetes Secret→Ingress · services · Kafka
Qubo IoT cloud, built solo
Hero Electronix · AWS, HiveMQ, MQTT, Kafka, Kinesis, Kubernetes, Terraform, Wowza
AWS bill −30% · broker +40%
THE PROBLEM
Qubo, Hero Group's line of consumer IoT devices, had no cloud platform. Devices needed secure, always-on connections, and their data needed somewhere to go, from an empty AWS account.
WHAT I DID
Designed the whole platform and took it live: VPC, IAM, Cognito, API Gateway and Lambda; a clustered HiveMQ broker (MQTT 3.1 and 5) with authentication, S3 discovery and JMX metrics; Kafka, Kinesis and Firehose pipelines; Kubernetes with blue-green releases; infrastructure as code in Terraform, CloudFormation and Ansible. Then ran it, fixing crypto and code-path issues in the broker and trimming spend.
IoT devices→MQTT/TLSHiveMQ cluster→Kafka · Kinesis · Firehose · SQS→DynamoDB · MongoDB · Cassandra · RDS · Redis
Shared CI/CD for 150+ engineers
adidas · Jenkins (Groovy), GitHub Actions, Checkmarx, Harbor, Flux
Monorepo build time −50%
THE PROBLEM
Many teams, many pipelines, each a little different. Releases were slower and less safe than they needed to be, and a Lerna monorepo took far too long to build and test.
WHAT I DID
Built and maintain the Jenkins shared libraries and GitHub Actions workflows that developers and QA release through, with Checkmarx SAST as a gate, images in Harbor and Flux for GitOps deploys to EKS. Reworked the monorepo build so build and test time came down by about half.
Bitbucket→Jenkins / GitHub Actions→Checkmarx SAST→Harbor→Flux→EKS · dev → prod
Cost, observability and AI for operations
adidas · OpenSearch, Prometheus, Grafana, Instana, Bedrock, MCP
K8s −15% · databases −20%
COST AND SIGNAL
Cut Kubernetes spend by ~15% with right-sizing, a cluster downscaler and weekend shutdown of non-production, and RDS, DocumentDB and ElastiCache spend by ~20%. Standardised one log format across frontend, backend and platform services in OpenSearch, so ops find and fix issues ~25% faster. Reworked alerts to cut noisy pages; set and report SLOs and MTTR.
AI IN OPERATIONS
Shared pipelines that take other teams' AI apps to production on AWS Bedrock using IRSA instead of long-lived keys. Internal MCP servers for AWS, AWS Knowledge Base, Atlassian and Instana so on-call engineers get context in one place. Bedrock RAG over runbooks and incident history, and a local stack on Ollama, FastMCP and ChromaDB.
On-call engineer→MCP servers
AWS · Atlassian · Instana→Bedrock RAG
IRSA, no static keys→Runbooks · incident history
Edge, CDN and API layer
Akamai · CloudFront · Kong
Cache and content rules in Akamai Control Center, certificate pinning at origin and edge, and WAF integration. CloudFront with TLS in front of S3, and Kong at the API layer.
Same provisioning, every time
Terraform · Service Catalog · IAM
Terraform modules and AWS Service Catalog products so new accounts and platform components come up the same way, with least-privilege IAM from day one.
Production databases
RDS · DocumentDB · ElastiCache
Day-to-day administration in production: parameter groups, backups, scaling and right-sized instance classes. At Qubo, tuned clustered MongoDB and Cassandra on EC2 for IoT ingest.
SmartOps, hackathon build
Jira · Confluence · search
Built at an internal AI hackathon: one search across Jira and Confluence, so engineers find related incidents and past fixes faster.
Kubernetes, the hard way
kops · kubespray · kubeadm
Created Kubernetes clusters on AWS before EKS was the default, managed Deployments, Services and Ingress with Jenkins X, and moved selected services from VMs to containers.
Live and recorded video
Wowza · HLS · MPEG-DASH
Tuned Wowza Streaming Engine for live and recorded HLS and MPEG-DASH streams for Qubo cameras, improving performance by ~10%.