Samskrutha Swamy

Sheet 00/Platform engineer/Berlin

Systems that hold up in production.

I'm Samskrutha, Sam for short, a platform engineer in Berlin. I design, build and run the infrastructure software depends on: cloud platforms, delivery pipelines, observability, and more recently the systems behind LLM products.

I help companies make that ground reliable, secure and affordable, and help engineers grow into this kind of work.

How I can help Get in touch
Berlin · · at a keyboard
Fig. 0 The author, hatched.
Years in software
12+
Years running platforms
7
Countries worked in
3
Cloud & Kubernetes certifications
5

Fig. 1 From idea to production. Click a station to see how I help there.

The life of a product feature Seven stations: idea, data, build, test, ship and observe left to right, and learn on the return path. Each station is a button that shows how Samskrutha helps at that step. worth it trusted right-sized gated boring idea data build test ship observe learn

Idea Pick the problem worth solving.

I help you decide what's worth building, agree what "good" looks like before anything is built, and settle which decisions stay with people.

Use-case review · success criteria · approval lines

Data One version of the facts.

Versioned data with contracts and freshness checks, so reports, services and language models all work from the same, current facts.

Data contracts · Postgres · SQL · retrieval & embeddings

Build The simplest design that works.

Choosing the simplest architecture that meets the need, including which model, if any. Clean interfaces, and tooling that makes your engineers faster.

Python · TypeScript · Anthropic & OpenAI APIs · MCP

Test Before a customer does.

Automated tests in the pipeline and, for LLM features, evaluation sets and grounding checks. Everything sits behind a release gate: if it fails, it doesn't ship.

CI gates · security scanning · LLM evaluations

Ship Run it on ground that holds.

Cloud, Kubernetes, secrets and CI/CD written as code, with rollouts that roll back in one command and the controls your auditors will ask about.

Terraform · Kubernetes (CKA) · Argo CD · GitHub Actions · SOC 2

Observe See problems before users do.

Metrics, logs and traces with consistent tagging, cost visible per service, and alerts that reach someone who can act, including the failure modes particular to LLM calls.

Datadog · Prometheus · Grafana · on-call

Learn Get better every week.

Every incident becomes a postmortem, a runbook entry and a better test, and your team ends up owning all of it without needing me.

Postmortems · runbooks · handover

Sheet 01/How I help

Problems I solve.

Situations I'm often brought in for, and what I do about them.

  1. 01

    “Our LLM feature works in the demo. We don't trust it in front of customers yet.”

    What I doI build evaluation sets and grounding checks so every answer is tested before release, and every number in it traces back to your own data.

    Release gatesFewer wrong answersProof for stakeholders

  2. 02

    “Things break in production, and we hear about it from customers.”

    What I doI set up observability that tells the truth, from services down to individual model calls, with alerts that reach the right person and an incident process the team can follow.

    Metrics, logs, tracesAlerts with ownersIncident process

  3. 03

    “The cloud bill keeps surprising us.”

    What I doI make spend visible per service and team, including model usage, add budgets and anomaly alerts, and cut what isn't pulling its weight.

    Cost per serviceBudgets & alertsRight-sizing

  4. 04

    “We want AI tools to use our internal systems, without opening a security hole.”

    What I doI build MCP servers and tool layers that give models scoped, read-only access by default, with authentication, isolation per customer and an audit trail.

    MCP serversLeast-privilege toolsAudit trail

  5. 05

    “Our engineers have coding assistants, but delivery hasn't sped up.”

    What I doI set up repositories, review gates and workflows so assistants do useful work within the team's existing process, then train the team to run it on their own.

    Repo conventionsReview gatesHands-on workshops

  6. 06

    “We're growing faster than our infrastructure, and an audit is coming.”

    What I doCloud, Kubernetes, CI/CD and observability written in Terraform, ready for SOC 2 and ISO 27001, and built so your team owns it after I leave.

    Terraform platformGitOpsAudit-ready controls

Sheet 02/Ways to work together

For companies, teams and people.

Pick what fits. Everything starts with a short call about what you're trying to do.

For organisations

A project, or a few months embedded.

A fixed piece of work with a clear finish line, or a stretch inside your team as the senior platform engineer you don't have yet.

  • A production platform, written in Terraform
  • Observability, on-call and incident process
  • Release testing for an LLM feature
  • SOC 2 and ISO 27001 groundwork

For engineering teams

Workshops on your own code.

One to three days on your real repositories, not slides, so the team leaves with a setup it understands and can run.

  • Terraform, Kubernetes or CI/CD practice
  • Observability and on-call that works
  • Coding assistants that fit your review process

For individuals

Mentoring, one to one.

For engineers moving into platform or DevOps work, or growing toward a senior role.

  • A learning path that fits where you are
  • Practice on real problems: Kubernetes, Terraform, on-call
  • Honest feedback on a CV, a design, or a move to Germany

The foundation underneath all of it

Terraform platforms on AWS & GCPKubernetes in production (CKA)CI/CD & GitOpsObservability & on-callIncident commandSOC 2 & ISO 27001FinOpsData contractsLeading teams

Tell me what you're working on

Sheet 03/Recent work

What I've been building.

What each team needed, and what I built.

2025 – now · Berlin

Bluefish AI

Platform, observability & data engineering

The goal. Show brands how AI assistants talk about them. Every day the platform asks ChatGPT, Claude, Gemini, Perplexity, Copilot, Meta AI and Google's AI Overviews about brands, then analyses the answers for sentiment, citations and accuracy.

  • Reliability for a multi-provider LLM pipeline. Empty completions fail loudly, provider timeouts are classified properly, and every channel has a coverage monitor so a silent gap gets noticed the same day.
  • Observability from nothing. Datadog APM, logs and unified tagging across every service, plus cost-anomaly alerts and log-to-metric work that cut the bill without losing a signal.
  • Cloud in Terraform, audit-ready. AWS with WAF, replication, dead-letter alarms and cost tagging, plus the infrastructure side of SOC 2 and ISO 27001.
  • Data products with contracts. Versioned data contracts (ODCS) and data products (ODPS), with a publisher CLI and compatibility checks before anything reaches a consumer.
  • Shared engineering tooling. Reusable Claude Code skills with their own tests, so the team works from the same playbook.

OpenAIClaudeGemini / Vertex AISageMakerAWSTerraformDatadogClaude Code skills

2024 · Lübeck

Perfood

Senior DevOps engineer · digital health

The goal. Move medical nutrition apps for migraine and diabetes onto a sovereign Google Cloud, where patient data and compliance aren't optional.

  • GKE from Terraform with Argo CD wired in.
  • Prometheus, Grafana, Loki and Thanos, with alerts cleaned up to the ones that matter.
  • Teleport and Vault for access, and the platform under an ML glucose-prediction service.

GCPGKEArgo CDVaultPrometheus

2019 – 2024 · Wolfsburg

Hexad for Volkswagen & CARIAD

Lead DevOps engineer · automotive

The goal. Give Volkswagen Group software teams a cloud they could trust with real traffic, and an operations team that answered when it mattered.

  • Led a DevOps team of five; operations lead and escalation point for CARIAD.
  • Fault-tolerant AWS and Kubernetes in Terraform, Jenkins with GitOps and security scanning.
  • Elastic logging on Kubernetes, Grafana and New Relic dashboards, and a Redshift Serverless migration.

AWSKubernetesTerraformJenkinsElastic

Sheet 04/Experience

Where I've done it.

Three countries, three continents, and a slow drift from writing the application to owning the ground it runs on. Click a stop.

Map of the route A dotted line connecting Mysore, Kalamazoo, Dallas, Wolfsburg and Berlin, in that order. Each stop is a button that shows that chapter. Mysore2008 Kalamazoo2012 Dallas2014 Wolfsburg2019 Berlin2024 →
Fig. 3 Not to scale. The Atlantic is wider than it looks.
5 / 5, Berlin

2024 – now · Berlin, Germany

Platform, observability and data.

A sovereign cloud for digital health at Perfood, then Bluefish AI: the platform, observability and data contracts behind a product built on large language models.

  1. 2025 –Bluefish AI, Berlin · Platform, observability & data engineering
  2. 2024Perfood, Lübeck · Senior DevOps engineer, digital health
  3. 2019 – 24Hexad, Wolfsburg · Lead DevOps engineer, Volkswagen & CARIAD
  4. 2014 – 19HCL, Dallas · Senior software engineer, finance & travel
  5. 2012 – 14Western Michigan University · M.S. Computer Science
  6. 2008 – 12Visvesvaraya Technological University · B.E. Computer Science

Sheet 05/Bill of materials

Toolbox.

Daily means I live in it, often means most months, when needed means I've shipped with it and can again.

LayerMaterials
CloudAWSGoogle CloudCloudflareAzure
OrchestrationKubernetesECSDockerHelmArgo CDFlux
Infrastructure as codeTerraformCloudFormationAnsible
DeliveryGitHub ActionsGitLab CIGitOpsJenkins
ObservabilityDatadogPrometheusGrafanaLokiElasticNew Relic
DataPostgres / RDSSQLData contractsStep FunctionsRedshift
SecuritySOC 2ISO 27001IAM / SSOZero-trust accessWAF
LLM & MLClaude CodeAnthropic APIOpenAI APICodexGemini / Vertex AIMCPLLM evalsEmbeddings & QdrantSageMakerLightGBM
LanguagesPythonShellTypeScriptJava
  • Certified Kubernetes Administrator
  • HashiCorp Terraform Associate
  • AWS Cloud Practitioner
  • Certified Scrum Product Owner
  • Azure Fundamentals

Sheet 06/Contact

Say hello.

A project, a role, or just a hard problem: write me a note. I read every one.

Send a note

Off the clock I'm with my wife and our two-year-old, or listening to a tech podcast while building something that doesn't need to exist yet.