I build the layerunderneath theproduct.
Infrastructure, automation, developer platforms, and the AI systems that sit on top of them.
What I can do
Infrastructure as something engineers consume rather than assemble. The measure is how little of it anyone else has to think about.
- Cloud infrastructure
- Internal developer platforms
- Infrastructure automation
- CI/CD
- Environment & deployment design
Social platformon AWS
I built and support the production infrastructure behind a social app with more than 80,000 users: containers, databases, deployments and monitoring, all defined as code.
- 80,000+
- AWS Mumbai, 2 zones
- 5 container services
- 4 AWS CDK stacks
0.03s
Fast, until it wasn't
Every service runs as a container on Fargate behind a load balancer, deployed by GitHub Actions through short-lived OIDC credentials. Typical API response time: 0.03s.
Ten times slower, with nothing on fire
Users reported the app had felt slow for two or three days. Response time had risen tenfold while the API's own servers sat at 7–18% CPU, so autoscaling never reacted.
Four workers doing one job
The background worker service had scaled itself from one copy to as many as four. Its scheduled jobs were never designed to run more than once at a time, so each copy repeated the same work against the cache.
A cache out of credits
The cache ran on a burstable instance. The duplicated work spent its CPU credits over three weeks until AWS throttled it to 20% of a core, and every request that touched the cache queued behind it.
Alarms that stayed green
A throttled machine shows low CPU, and an average hides a slow tail. The existing alarms watched exactly those two things.
Remove the cause, then the blind spot
Pinned the worker to one copy. Moved the cache to a fixed-performance instance. Added alarms on CPU credit balance and 90th-percentile latency, about two days of warning. Handed the team the code change needed before the worker can safely scale again.
Back to 0.03s
Typical API response time returned to 0.03s, and the next time credits start draining, someone gets told.
- ECS Fargate
- RDS PostgreSQL
- ElastiCache Redis
- Amazon Keyspaces
- CloudFront
- AWS WAF
- CloudWatch
- AWS CDK
- GitHub Actions
More work
- zaneatlas.comA technology studio's site: fast on phones, readable with or without animation, deployed automatically, and built so every project inquiry is routed to the right place.Lighthouse lab scores, October 2026: 97–100 performance, 100 accessibility.
- BackedHireAI-assisted resume screening that assesses each candidate against the criteria that matter for the role and shows the evidence behind every assessment, so people stay in charge of the decision.Every assessment can be explained, reviewed and challenged by a person.
How I got here
AR Weather
Putting data where the weather actually is
A weather app you pointed at the sky. The interesting problem was anchoring remote data to the place you were standing.
Facial Recognition Attendance
Removing a ritual nobody defended
What happens when a model meets a workflow that real people have to live inside.
Music on a Distributed Ledger
Learning a technology by arguing with it
A music app on blockchain, mostly to learn where the model stops making sense.
Windfarm Digital Twin
Software that has to agree with reality
A live software model of physical machines. Correctness measured against the world rather than a spec.
Test Automation for SuccessFactors & UKG
Making a repetitive process into an asset
Automated testing across two enterprise HR platforms, reliable enough that people stopped re-checking by hand.
Copilot × Jira
Closing the gap between the ticket and the branch
AI assistance during development that knows which work item it's actually serving.
Supply Chain Operations Agent
Where an agent may decide, and where it may not
An agent in a domain of exceptions and partial information. The question was where its authority should end.
Self-Service Developer Platform
Building the floor other people build on
Paved paths for provisioning, deploying and running services. Where my interest moved to the layer underneath.
- See the current work.
What I'mlearning
MSc Computer Science
University College Dublin · Negotiated Learning
Formalising the parts I picked up by building, and aiming the coursework at the areas I already work in.
BTech Computer Science and Technology
Dayananda Sagar University, Bengaluru
Where the first experiments came from, and where I found out the unfamiliar technology is rarely the hard part.
How far in
Filter by depth. Three squares is where most of the work is; one is just opening up.
- Platform engineeringThe centre of the work right now.
- AI engineeringBuilding with models rather than around them.
- Agentic systemsSpecifically: bounding what they're allowed to do.
- Cloud infrastructure
- Developer platforms
- Infrastructure automation
- AI-assisted development
- System architecture
- Developer experience
- Modern web developmentThis site is part of that.
- Design systemsSo is the brand it's built on.
- Building productsThe longer-term direction.
Questions I don't have answers to yet
Where should an agent's authority end, and who decides: the platform or the team using it?
The personbehind it
- An MSc at UCD, alongside the work: infrastructure, automation and the AI systems on top of them.
- Experiments in AR, computer vision and distributed ledgers at university. Then enterprise work: a windfarm digital twin at HCL and test automation at ANKO.
- Where agents belong in operational workflows, and what developer tooling looks like when AI is assumed rather than added.
What isn't the job
Have something thatdoesn't have anobvious answer?
I'm most useful early, when the problem is still being described, the architecture is undecided, and it isn't obvious whether the thing should be built at all.