AWS cost and Kubernetes
Right-size what you run with Karpenter, Spot only where it's safe, and upgrades off extended support. No migration.
HomeEnterprise
Sivaro Enterprise · cloud · data · AI
How do you explain “My AI miscalculated”?
There is no danger that Titanic will sink.1912is the new
Apologies for missing that.2026
AI cannot sign. We can.
Cloud, data and AI engineering for enterprises: ClickHouse, Kubernetes and Karpenter, AWS cost, and in-house cloud. Senior engineers who put their name to the work, with read-only access first.
Diagnostic · sign-offacme-prod · AWS · read-only
Sample · not a clientSigned by the engineer who ran the testsName · date
Sivaro Softwares LLPRegistered in India, 2024
HyderabadOperating office
NDA firstBefore any technical call
Read-only firstNo write access to start
Priced firstQuoted before work starts
15+ recurring clients, includingVolzaFloqerMinistry of Power, IndiaSyndie.ioScaleGlobalDigital Align IncDigital Brinks
What we do for enterprises
Six areas. The same standard in each: test first, then fix.
Data platforms and analytics
Real-time analytics, pipelines and migrations, including ClickHouse in production.
Production AI
RAG, agents and evals that cite their sources, with a cost per answer you can track. Sivaro AI →
Call scoring and coaching
Every call transcribed, scored and coached, with personal data masked. Built for Volza →
Platform engineering
Self-service deploys with policy checks, default-deny APIs and an audit trail for every change.
In-house cloud
When renting stops making sense: colocation or bare metal, planned on your own numbers, with exit waivers where they apply.
Case studies
Systems in production. Numbers once clients sign them off.
Call Center OS for Volza
Every sales call tracked, scored and coached.
- Challenge
- Managers could listen to a handful of calls, so coaching stayed generic and patterns across Volza's sales team went unseen.
- What we built
- A queue-based pipeline: transcription and speaker separation, then sentiment, close likelihood, talk ratio and a rubric score in parallel, feeding coaching views and a leads board.
- Why it holds up
- Personal data masked in transcript and audio. Retries and a dead-letter queue, so one failing analysis never blocks a call. Installs with one command.
ClickHouse Toolkit
Three tools for running ClickHouse in production.
- Challenge
- ClickHouse schema changes can't be rolled back inside a transaction, and a slow cluster gives no quick view of what's running.
- What we built
- Versioned up-and-down migrations recorded inside ClickHouse, a live query monitor, and runbooks with 18 system queries written after real incidents.
- Why it holds up
- Re-runs are safe and every change can be undone: each migration is stored with its rollback SQL and a checksum.
Face recognition on live cameras
Prototype delivered to the client.
- Challenge
- Security cameras record all day, but nobody can watch every feed live or search it afterwards for who appeared and when.
- What we built
- Reads an IP camera feed once a second, detects faces with InsightFace, matches them against known people, enrols new ones and logs every sighting with a photo to a live dashboard.
- Why it holds up
- Tuned for night and traffic cameras: every detection passes size, shape, brightness, sharpness and confidence checks, and the stream reconnects itself when a camera drops.
English speaking coach
Deployed for the client.
- Challenge
- Learners read English well but freeze when they have to speak it, and tutors who give feedback on speaking are expensive.
- What we built
- Read-aloud practice scored word by word, AI role-play on ten everyday topics at three levels, daily quizzes, streaks and a weekly leaderboard.
- Why it holds up
- 109 automated tests gate every deploy, and a release only counts once the database answers a health check.
How an engagement runs
Five steps, each with a document you can take to your leadership.
- 1
NDA and scoping call
30 minutes with an engineer. Is it real, what's it worth, who decides by when.
Free - 2
Diagnostic, read-only
Prescribed tests on your own numbers, every alternative ruled in or out, and one signed page of findings.
Priced by the number of tests - 3
Build, with weekly demos
The first week ends with software you can open, and so does every week after.
Fixed price · 50% upfront - 4
Run and report
A one-page monthly ledger: cost per workload, what changed, and what's next.
Monthly - 5
Hand over
Runbooks your team can run without us, documentation, and our access removed.
Included
Security and data handling
What we touch, and what we never do.
Your security team gets the exact list of what we read, before we read it.
How we work
- Read-only access first, listed resource by resource
- Or a script you run yourself, sharing nothing with us
- Least-privilege roles, scoped to the engagement
- Data stays in your cloud and your accounts
- Access removed at handover
What we never do
- Write, restart, scale or delete during a diagnostic
- Need your prompts to price your AI answers
- Move a model or a workload without your sign-off
- Take referral fees from vendors we choose
SOC 2 and ISO 27001 ready. We don't just create and leave. We constantly fix and improve too.
For procurement
The facts your vendor form asks for.
Anything missing, ask on the first call and we'll send it in writing.
| Legal entity | Sivaro Softwares LLP |
|---|---|
| Operating office | Hyderabad, India |
| Founded | 2024 |
| Contracts | Mutual NDA, then an MSA with a statement of work per project. We can work on your paper. |
| Pricing | The diagnostic is priced by the number of tests it needs. Builds are a fixed price, agreed before work starts, 50% upfront. |
| Invoicing | US dollars, with the rupee amount beside it, or at a fixed conversion rate agreed in the contract |
| Tax registration | GSTIN 09AFIFS3854D1ZS |
| Certifications | SOC 2 and ISO 27001 ready. Not yet audited. |
| IP ownership | Clients own the KPIs. We own the implementation. |
| Contact | [email protected]. We reply within an hour. |
Start here
The diagnostic. Read-only, with prescribed tests. Fix what you run first; leave AWS only as a last resort.
A read-only role or a billing export is all it needs. We list the tests before we start: every cluster with its version and support dates, idle capacity, traffic between zones, leftovers, your ten largest cost lines and your next commitment renewal. The price follows from how many tests your setup needs.
1 · priced by testsDiagnostic
Read-only, or run our script and send us nothing.
2 · fixed price · weeksRight-size in place
Karpenter, Spot where safe, off extended support, less zone traffic.
3 · scoped projectRe-platform one piece
When one line dominates the bill, such as analytics to ClickHouse.
4 · only if the numbers say soIn-house cloud
Colocation or bare metal, with exit egress waivers where they apply.
We learn from the best · what these teams published
20%off Tinybird's overall AWS bill after moving to EKS, Karpenter and SpotTinybird blog, June 2025
12%off Slack's EKS compute cost with KarpenterAWS Containers Blog, May 2024
40%below on-demand for Anthropic, year over year, with flexible instance types and Spot through Karpenterre:Invent 2023 talk, as reported by The Stack
5%saved in FY2026 across Salesforce's fleet of 1,000+ EKS clustersAWS Architecture Blog, January 2026
The largest fleet saw the smallest share, which is why we test yours on your own bill and never promise a percentage.
Questions
What enterprise teams ask first.
Do we have to leave AWS?
No, and most teams shouldn't. We start by making what you run cheaper, and suggest leaving only when your own numbers say so.
What access do you need?
Read-only access or a billing export. The diagnostic lists every resource it reads. You can also run our script and share nothing.
Can you work on our contract and security paperwork?
Yes. Mutual NDA first, then your MSA or ours, with a statement of work per project. Send your vendor questionnaire and we'll answer it in writing.
Who works on our account?
Senior engineers only. The people who scope the work are the people who build it. Nothing is handed down a layer.
Can you promise a saving?
No. We show it on your own numbers, before and after.
What does it cost?
The scoping call is free. The diagnostic is priced by the number of tests your setup needs. Builds are a fixed price, agreed before we start, 50% upfront.
How long does it take?
As long as the tests take. You get the list of tests, the timeline and the price before the diagnostic starts.
Can't our team do this with AI?
An AI can suggest a fix. It can't be accountable for it. Our engineers run the tests on your numbers and sign the findings.