cloudgeist.cloud

BotondGeiszt

Cloud Engineer · AI Integrator · Budapest

botond@cloudgeist:~$ terraform plan -target=module.botond
  # module.botond.contract.dach_remote will be created
  + resource "contract" "dach_remote" {
      + location     = "Budapest, HU · CET"
      + mode         = "remote / hybrid"
      + availability = "immediate"
      + focus        = ["aws", "eks", "finops",
                       "ci-cd", "llm-infra"]
      + data_region  = "eu"
      + lock_in      = false
    }
Plan: 1 to add, 0 to change, 0 to destroy.
 
Type help for commands, or tap one below.
try

Cloud-native infrastructure and AI agents in production. I build AWS and Kubernetes platforms for DACH teams, and I find the money those platforms quietly waste.

Budapest, CET. Open to remote contracts in Germany, Austria and Switzerland.

  • 4+ yrs production DevOps
  • 20+ microservices on one CI/CD platform
  • 5 open-source reference repos

What is your AWS bill leaking?

Accounts without a FinOps owner typically lose 15–30% to idle compute, orphaned storage and oversized instances.

€80,000 €6,667 / month
Typical waste
€1,000–€2,000 per month
€12,000–€24,000 a year
Audit fee at this estimate
€1,200–€2,400excl. VAT
10% of the waste the audit identifies in a year, minimum €1,500.

Report within 10 working days of read-only access. You send one email; I reply with times for a 15-minute intro call.

An estimate from the stated 15–30% range, not a quote. The audit replaces it with per-finding numbers from your account.

How the audit works

  1. You send one email. The draft carries your estimate.
  2. A 15-minute intro call. What you're running, where it hurts, and whether I can help.
  3. A read-only, fixed-scope audit. Your billing data stays in your account.
  4. The report. JSON findings, a ranked Markdown report, and the top five fixes with exact commands.

What the audit checks

Five services, each with open-source code you can read before we talk.

Senior DevOps and cloud engineer, available for remote contracts in the DACH market. Production cloud platforms, and the cost discipline that keeps them sane.

  • aws

    AWS platform engineering

    Multi-account landing zones, networking, IAM, and security baselines. Governed foundations before the technical debt compounds.

    aws-platform-baseline
  • k8s

    Kubernetes / EKS platforms

    Cluster provisioning, GitOps delivery, autoscaling, observability, and policy enforcement. Production EKS without the 2AM incidents.

    eks-platform-reference
  • finops

    FinOps and cloud cost

    Find and fix waste, then set up the alerting and governance so it doesn't come back. Often pays for itself in the first month.

    aws-finops-toolkit
  • ci/cd

    CI/CD and DevSecOps

    Reusable pipelines with security gates, approval flows, and OIDC auth, so no long-lived credentials sit in CI.

    cicd-pipeline-templates
  • llm

    LLM / AI infrastructure

    API cost controls and self-hosted model deployment on AWS, from a managed API proxy to self-hosted GPU on EKS.

    llm-infra-aws
DACH-ready

CET timezone, fluent English, EU-based invoicing.

GDPR and data residency

EU regions by default and least-privilege access. Your data stays yours.

Partnership over projects

Sustainable systems you can run after I'm gone, with no lock-in.

What the audit checks, and what you get back.

Built for SaaS teams with €20k–200k a year in AWS spend and no dedicated FinOps function.

AreaWhat I look for
ComputeIdle EC2 (<5% CPU over 7 days), oversized instance types, unused Elastic IPs
StorageUnattached EBS volumes, S3 buckets without lifecycle rules, stale snapshots
NetworkingUnderutilised NAT Gateways, idle load balancers
DatabaseRDS instances under 10% CPU, unencrypted snapshots
SpendAnomalies against your prior 30-day baseline
What you get back
  • JSON findings, machine-readable, drop straight into CI.
  • Markdown report with estimated monthly saving per finding, ranked by impact.
  • Remediation narrative: the top five actions in plain English, with the exact commands.
The fee

10% of the identified yearly waste, excl. VAT

Identified waste is the yearly saving of every finding validated in the report, at your current AWS prices.

Minimum fee

€1,500 excl. VAT

Applies when 10% of the identified waste comes to less. Read-only access, fixed scope, report within 10 working days.

Top findings · estimated savings €1,840/month

Example report. The first 2 findings are shown; the total covers the full report.

  1. Three unattached EBS volumes (gp2, 500GB total), €47/month. Not attached to any instance and untouched for 45+ days. Safe to snapshot and delete unless tied to a manual backup process.
  2. NAT Gateway eu-west-1a, €127/month, 0.2GB/day throughput. Carrying almost no traffic, so it is a candidate for consolidation.

Have me run it

I run the analysis, validate every finding against your context, and deliver the report with a prioritised remediation plan.

Request the audit

Run it yourself

The toolkit is open source and uses a read-only IAM policy. No data ever leaves your account.

aws-finops-toolkit on GitHub

From telecom scale to AI in production.

Four years running CI/CD and Kubernetes for a cloud-native IP Multimedia Subsystem serving millions of users, then cloud migration and LLM systems. Select a role to read it, or open the full CV with stack, certifications and education.

Bars are to scale. A dashed outline marks education; the amber line is today.

    Notes from shipping cloud and AI systems.

    Small, specific things learned on real projects, the opposite of a tutorial. Also available as an RSS feed.

    • S3 says it worked. Check anyway.

      Five times S3 or its tooling gave me a signal that looked like proof and wasn't, and the second check that caught each one.

      #aws
    • The build that remembered its mistakes

      Seven weeks of AI-driven AWS work got better not from a bigger model but from three kinds of memory — and the scars mattered most.

      #ai #claude-code #reflection
    • Vault was the easy part

      An Obsidian-style knowledge vault doesn't survive daily work on its own — it needs skills wrapped around it to stay alive.

      #claude-code #reflection
    • Deep water

      Three AI systems shipped via cognitive collab with Claude Code, and the meta-lesson I only saw by stopping to look back.

      #ai #claude-code #reflection
    • Hello, world

      Why I'm starting to write down what I learn — and what to expect on this blog.

      #meta

    Each S3 signal that lied, and the second check that caught it

    The second check must see the state some other way, so it can't share the first one's bug.

    Looked like proofSecond check
    delete-bucketlist-buckets, polled until head-bucket agrees
    delete-objectslist-object-versions
    --query length()--output json | jq length
    detach-user-policylist-attached-user-policies AND list-attached-group-policies
    put-bucket-policya real request from a principal you didn't enumerate
    Read the post

    Three memories for a long agent build

    Seven weeks, one AWS proof of concept, 122 pull requests, driven end to end through an AI agent.

    WHYThe vault: architecture and decisions, provenance-anchored. The layer that survives handover.
    HOWA live code graph: functions, routes, call chains, queried before reading files.
    SCARSOne-fact files of the agent's own mistakes, reloaded every session. 61 by the end.
    The model does not get smarter across a project. What compounds is memory.
    Read the post