Services · technical overview

Three clearly defined service areas
with a technical focus.

On this page you will find technical details on each of our three disciplines. We are happy to put the technical terms into context in a conversation.

01 - SaaS

.NET + PostgreSQL,
packed into Docker.

We build production-ready web platforms on a stack that has been stable for years and will still receive updates in five years. Frontend in Next.js, backend in ASP.NET Core, data in PostgreSQL. Containerised, automatically tested and shipped through separate environments.

What you get

→Frontend: Next.js (App Router), TypeScript, Tailwind or CSS modules
→Backend: ASP.NET Core 8 with minimal APIs or MVC, EF Core
→Database: PostgreSQL with migrations, seed data, PITR backups
→Containers: multi-stage Dockerfiles, docker-compose for local setup
→CI/CD: Azure DevOps pipelines - build, test, scan, automated deployments following defined rules
→3 environments: dev → sandbox → production, with an E2E gate before prod
→Authentication: Keycloak SSO (OIDC, SAML). Integrates with an existing Active Directory or Entra ID
→Provisioning: Ansible playbooks for reproducible server setups
→Cloud: Azure (DE West Central / North Europe) or on premise
→Monitoring: OpenTelemetry, logs, metrics, traces - in Grafana on request
→Backup tests: not just taking backups, but restoring them automatically

● 3-stage deployment

Development

Fast feedback through automated builds and deployments.

Unit + integration

→

Sandbox

Close to production. This is where fully automated E2E tests run.

Unit + integration + E2E

→

Production

Only after a green sandbox run.

Smoke + health checks

Promotion from sandbox → production only when all E2E suites are green. A roll-back path is available at all times.

● azure-pipelines.ymlYAML

trigger:
  branches: [main]

stages:
- stage: Build
  jobs:
    - job: BuildAndTest
      steps:
        - task: DotNetCoreCLI@2  # build + unit tests
        - task: Docker@2         # multi-stage image
        - task: AquaSecurity     # trivy scan

- stage: Sandbox
  dependsOn: Build
  jobs:
    - deployment: ToSandbox
      strategy:
        runOnce:
          deploy:
            steps:
              - script: ansible-playbook deploy.yml
              - script: npx playwright test  # E2E

- stage: Production
  dependsOn: Sandbox    # only on green E2E tests
  jobs:
    - deployment: BlueGreen
      environment: prod-approval

Performance target

< 200 ms p95 API latency

Target value for typical business workloads on appropriately sized Azure infrastructure.

Distribution channels

Steam·Microsoft Store·Own shop + Stripe·Direct download

We set up whatever suits your audience - from store models to your own checkout with invoicing and SEPA.

02 - Desktop

Electron + Next.js,
signed, packaged and ready for updates.

For applications that need to work offline, be installed locally or be sold through stores, we build with Electron and Next.js. Web developers know the frontend, and end users know the delivery from any standard installer.

What we deliver

→Code signing for Windows (EV) and macOS (Apple notarization)
→Auto-update with Squirrel or electron-updater - differential
→NSIS / MSI / DMG / AppImage installer packages
→Steamworks integration: achievements, cloud saves, Workshop
→Microsoft Store: MSIX packaging & submission pipeline
→Your own shop: Stripe checkout, licence server, licence keys
→Native modules where needed (audio, video, hardware access)
→Crash reporting via Sentry with source maps

+ AI / ML - cross-cutting

AI integration
for concrete use cases.

AI is not a focus area of its own but a capability we build into SaaS and desktop projects wherever it makes sense. From integrating existing model APIs to custom workflows on company data. With an eye on data protection, cost control and the ability to swap providers.

What we do

→LLM integration. OpenAI, Anthropic, Mistral, Llama - also on premise via Ollama or vLLM
→RAG & knowledge bases. Documents, wikis, databases - vectorised with pgvector or Qdrant
→Agents & workflows. Tool calling, structured outputs, multi-step tasks with guardrails
→Image generation. Stable Diffusion, Flux, ControlNet - hosted or API based
→Audio & speech. Whisper for transcription, TTS voices, voice cloning where required
→Fine-tuning & training. LoRA adapters, domain models, classic ML with scikit-learn / PyTorch
→Evaluation. Evaluation with test sets, A/B comparisons and quality metrics for model responses.
→EU hosting. Models on Azure DE / North Europe or on premise, where this is required.

● All-round AI competence

LLM / text100%

Vision / image gen100%

Speech / audio100%

Classic ML100%

Our focus is not on fundamental research but on usable AI integration in production software.

Plug-inIntegrate existing models

Put GPT, Claude or Gemini behind your app - with caching, rate limits and fallbacks. 1-4 weeks.

BuildYour own AI workflow

RAG, agents or vision pipelines built specifically for your use case. 4-12 weeks.

TrainFine-tuning on your data

LoRA adapters or classic ML models when off-the-shelf is not enough. 6-16 weeks.

03 - Product security

We treat product security as a service area in its own right.

Threat modelling, ISO 27001 preparation and secure SDLC coaching - implemented hands-on for teams that want to approach security requirements in a structured way.

Go to the security page

● ISO 27001 ready

Let us talk

One idea. One coffee.
One clear next step.

30 minutes, no strings attached - in German or English, on site in NRW or remote. We listen, give you an honest assessment, and sometimes say: “Don't build that.”

Arrange a meeting

info@huelake.com

Three clearly defined service areaswith a technical focus.

.NET + PostgreSQL,packed into Docker.