Public sector · Western Europe

European Municipal Services Cloud + AI enablement

Modernized a legacy application stack to cloud infrastructure and deployed a citizen-facing AI assistant for common service requests.

Timeline
9 months · phased rollout across 6 service lines
Team
1 architect, 2 cloud engineers, 2 backend, 1 AI engineer, 1 change-management lead
Focus
Cloud + AI enablement

The problem

  • A monolithic .NET application on aging on-prem VMs handled 12 citizen services with frequent outages during peak periods.
  • Frontline staff manually triaged ~1,800 inbound emails and forms per week, routing them across 9 departments.
  • Strict GDPR and data-residency requirements blocked off-the-shelf SaaS chatbots.

The solution

  • Strangler-fig migration: kept the legacy app behind an API gateway while progressively moving service lines to containerized services in an EU-region cloud.
  • Citizen-facing assistant trained on official policy documents, FAQs, and form metadata — answers in the local language and routes complex cases to the correct department.
  • Internal triage copilot for staff that classifies inbound requests, pre-fills case forms, and suggests next actions based on prior cases.

Architecture highlights

Cloud foundation

EU-region Kubernetes cluster with IaC (Terraform), private networking, VPN to on-prem for legacy integrations.

Data residency

All citizen data + LLM inference inside the EU region; self-hosted embedding model + EU-hosted LLM endpoint for sensitive flows.

AI layer

RAG over versioned policy corpus with citation-required answers; classifier model fine-tuned on 18 months of historical case data.

Identity

Integration with national eID for authenticated citizen flows; role-based access for staff with full audit trail.

Observability

Centralized logs, traces, and PII-scrubbed prompt logs for compliance review.

KubernetesTerraformPostgreSQLEU-hosted LLMOpenTelemetryKeycloak

Measurable outcomes

Indicative figures from the engagement. Exact numbers are placeholders pending client approval to publish.

~45%
Lower infra cost
vs. previous on-prem + licensing baseline.
~70%
Auto-routed requests
Triage copilot adoption across 6 service lines.
0
Data-residency exceptions
All inference and storage inside EU region.
Working on something similar?

Let's scope a pilot or a discovery sprint.

Start a conversation