Platform Architecture & Kubernetes
Build fleets that scale on demand, IAM that developers can provision themselves, and CI runners that keep teams isolated from each other. The kind of platform where filing a ticket stops being the default answer.
Lead Platform Engineer · Roche Information Solutions · Pune, India
Lead Platform Engineer with 7+ years building the internal developer platforms that engineering teams ship on — across AWS and GCP, at the scale of 500+ enterprise accounts. Currently leading a 5–7 person team at Roche, writing code every day while owning the product roadmap and keeping stakeholders aligned.
Open to new opportunities
About
The idea I keep coming back to: if an engineer has to remember a rule, the platform has already failed. Compliance should be built in, not chased after the fact.
I've spent 7+ years building the platforms other engineers build on — CI fleets, OS image pipelines, self-service IAM, and the AI tools that make those systems easier to live with. Right now I lead a 5–7 person team at Roche, running infrastructure behind 500+ enterprise accounts. I still write code every day.
Core Domains
Build fleets that scale on demand, IAM that developers can provision themselves, and CI runners that keep teams isolated from each other. The kind of platform where filing a ticket stops being the default answer.
Track down the defaults nobody changed, the images nobody cleaned up, and the instances running at the wrong size. Turned those into $40K+ in annual savings at Roche without asking anyone to change how they work.
GitHub Actions, GitLab CI, ArgoCD — built from scratch or rebuilt when scale broke the original. The goal is always the same: pipelines that run fast, fail clearly, and don't wake anyone up at 3 a.m.
Terraform, Terragrunt and Helm for infrastructure that actually matches what's in the code. Packer and Ansible for OS images that are hardened before they boot — CIS benchmarks and SBOM generation included.
Short-lived credentials over long-lived keys. SCPs and permission boundaries that make the wrong thing hard to do, not just against policy. Compliance built into the platform so engineers inherit it on day one — I've taken this work through ISO-27001 audits.
AI tools that reduce real toil — a debugging CLI that cuts root-cause analysis from hours to minutes, a code-review agent that asks better questions, and cost controls so the spend bill never surprises anyone. Built on Amazon Bedrock and open-weight models.
Impact
A week → A day
patch cycle for 1,000+ instancesHours → Minutes
root-cause analysis on a failing deployTicket queues → Self-service
IAM role access for developersChased → Inherited
compliance posture across the estateOwnership
1,000+ instances a month across 100 AWS accounts, on a schedule nobody files a ticket for.
5 OS families and 20+ compute variants, hardened to CIS benchmarks before anyone boots them.
1,000+ CI jobs a week across 10 tenants, isolated from each other by default.
Track record
Lead Platform Engineer | DevOps Lead · Pune
Lead a team of 5–7 engineers spread across time zones. We grew from an AWS-only setup into a full multi-cloud platform serving three major business groups. I own the roadmap and run the quarterly planning, represent the team in ISO-27001 audits, and still write code and review pull requests every week.
DevOps Engineer | Primary On-Call Engineer · Hyderabad
Service owner and primary on-call engineer. Owned the incident response process — formal root-cause analyses, preventative fixes, and the runbooks that stopped the same issues recurring. Also tackled two larger engineering problems: scaling a data pipeline to process terabytes of data, and planning a zero-downtime load balancer migration.
AWS DevOps Engineer · Nagpur
Built and owned Jenkins CI/CD pipelines for a large microservices project. Automated disaster recovery on AWS, and helped migrate the infrastructure from on-premises to the cloud using Terraform and CloudFormation.
Open source
Credit-risk signals are opaque for new-to-credit underwriting and behavioural auditing.
An explainable-AI diagnostic framework with SHAP-based insights, event-driven design and a Streamlit front end.
Every AI task hitting the largest available model wastes money on questions a smaller one answers fine.
An autonomous routing agent that picks the cheapest capable model at runtime, with Ollama and GGUF local-model support.
Choosing a foundation model without comparable cost and latency evidence is guesswork.
A CLI that runs one prompt across selected Bedrock models and writes a cheapest-first Markdown report.
Writing
One default setting, left untouched, quietly turning into a five-figure quarterly bill — and how to catch it before finance does.
Every pod you have ever run has an extra container you did not declare. What the pause container does, and why it matters when things go wrong.
Sessions, clients, resources and cross-account assume-role — the mental model that makes boto3 credential errors stop being mysterious.
Verified
Contact
If you're hiring for a Lead Platform Engineer or engineering team lead and want someone who is still hands-on — feel free to reach out.