contact
Tell me what’s on fire
Book a free 30-minute diagnostic call, or send a message. On the call we look at the actual problem — your cluster, your gateway config, your bill, your eval output — and you leave with a working diagnosis and a rough shape for the fix, whether or not you hire me. I reply to messages within 24 hours on weekdays. You talk to me directly, never a sales team or a bench.
Remote-first, based in India (IST), working across IST, CET, and US timezones with teams in India, Europe, the US, and APAC. Not sure which bucket your problem falls into? Start with the consulting services, read what similar engagements looked like, or skim the production write-ups in Insights.
Send a message
The more context you give me about your stack and the symptoms, the faster I can help. I reply within 24 hours.
Or grab a time directly
A free 30-minute diagnostic call.
after you book
What happens on the first call
The first call is a diagnostic, not a pitch. It is free, it is 30 minutes, and it is the same conversation whether you end up hiring me or not.
- 01
You book, I read up
Pick a 30-minute slot. Before we talk I read whatever you sent — error text, dashboards, a repo, a bill breakdown — so the call starts at the problem instead of at introductions.
- 02
Thirty minutes on the actual problem
We screen-share your cluster, your gateway config, your cost explorer, or your eval output. I ask questions, form a hypothesis out loud, and tell you what I think is wrong and what I would rule out first.
- 03
You leave with a read and a shape
By the end you have a working diagnosis, the next two or three things to check, and a rough shape for the work: emergency fix, scoped project, or ongoing advisory. No obligation to hire me for any of it.
- 04
If we work together, a written scope
I send a short written scope — deliverables, timeline, how we work, and what would make me tell you to stop. We start once you agree to it. You always work with me directly, never a bench.
Bring this to the call
- ✓The real error text or status.
kubectl describe podandkubectl logs --previousbeat any summary of them. - ✓For a bill: a Cost Explorer or billing export broken down by service and environment.
- ✓For an AI system: five to ten real inputs where the output was wrong, plus what the right answer was.
- ✓What changed recently. Deploys, node upgrades, chart bumps, model swaps, traffic shifts.
- ✓Who can approve access and decisions, and what your actual deadline is.
What I don’t take on
- ✕Frontend, mobile, and general product engineering. Not my lane, and you can hire better.
- ✕Pure data engineering, BI, and analytics work with no infrastructure or model-serving component.
- ✕Crypto and blockchain infrastructure.
- ✕Full-time employment. I am available for contract, fractional, and staff-augmentation work only.
- ✕Engagements with no decision-maker and no path to access. They burn budget and ship nothing.
how work gets scoped
Four engagement shapes
Almost everything I do fits one of these four shapes. We pick the shape on the first call, based on how urgent the problem is and how well-defined the outcome is. Scope, milestones, and price go in writing before any work starts.
| Shape | Use it when | Typical scope |
|---|---|---|
| Incident call | Production is down or degraded right now | Hours. Live debugging on a call, then a written note on the root cause and the guardrail that stops a repeat. |
| Scoped project | A defined outcome: a cost pass, a gateway build, a RAG system, a migration | One to six weeks, fixed scope. Milestones agreed up front, handover docs and runbooks at the end. |
| Fractional / retainer | You need senior infra or AI engineering judgment on tap, not a full hire | A recurring block of days per month. Architecture review, on-call escalation, and unblocking your team. |
| Staff augmentation | A team is short a senior engineer for a quarter | Embedded with your team on your tooling and your standup, for an agreed number of days per week. |
Common starting points: infrastructure debugging and incident response for a cluster that is down, cloud cost optimization for a bill that grew faster than traffic, API gateways and networking for ingress and mesh work, and RAG systems and AI chatbots for an assistant that demos well and fails in production.
Questions before you book
How long is a typical engagement?+
It ranges from a single incident call measured in hours to a quarter-long embedded engagement. Most work lands in the one-to-six-week band: a cloud cost pass, a gateway or ingress rebuild, a production RAG system, or a Kubernetes migration. I would rather scope something small, finish it, and let you decide whether to continue than sell you a six-month commitment on day one.
How do you charge — hourly, daily, or fixed price?+
Day rate for advisory, retainer, and staff-augmentation work; fixed price for scoped projects where the deliverable is clear enough to price. Incident calls are billed by the hour. I quote after the first call, once I know what the problem actually is, and the quote is in the written scope before any work starts. I do not charge for the first 30-minute call.
Will you sign an NDA or work under our contract?+
Yes to both. Most of my work is under NDA, which is why the case studies on this site name outcomes rather than clients. I am happy to sign your NDA before the first call if you need to share specifics to describe the problem, and I work under client paper or my own, whichever your procurement prefers.
What timezone do you work in and how fast do you reply?+
I am based in India and work in IST, which overlaps comfortably with European mornings and US mornings on either side of the day. I reply to email and the form on this page within 24 hours on weekdays. For active incidents, book the earliest call slot and say "incident" in the message — those get pulled forward.
What should I send before the call?+
Whatever the problem actually looks like. For Kubernetes: the output of kubectl describe and kubectl logs --previous, and the manifest. For cost: a Cost Explorer or billing export screenshot by service. For an AI system: a handful of real failing inputs and the expected outputs. Raw artifacts are more useful than a written summary of them.
What kind of work do you turn down?+
Anything where I would be the wrong person: frontend and mobile product work, pure data engineering and analytics, crypto, and long-term full-time employment. I also turn down engagements where nobody on your side can make decisions or give access, because those burn budget without shipping anything.