The problem

• 14 separate citizen-facing applications spanning 4 government departments, each with its own login, branding and case-tracking model.
• Median citizen wait time for cross-department cases (e.g. permits requiring multiple approvals) was 47 days; 9% of cases lost between systems.
• Multi-language requirement (Russian / English / Hebrew) was implemented inconsistently — three departments had translations, the others did not.
• No unified audit trail. FOIA and parliamentary inquiries required manual reconciliation across systems.
• Mandate to deliver consolidated portal within 24 months, with phased citizen migration.

The solution

• Single canonical citizen identity: federated identity layer mapping each legacy account to a unified citizen profile, with consent-based migration.
• Case orchestration spine: .NET 10 workflow engine routing cases across departments based on declarative rules. Legacy systems remained as adapters during phased migration.
• AI-powered case routing: LLM classifier suggests the correct department + form for free-text citizen submissions. Staff review and confirm; the suggestion is advisory, never autonomous.
• Multi-language pipeline: all case content translated and indexed in Russian, English and Hebrew at ingest. Citizens correspond in their language; staff see translations side-by-side with originals.
• Unified audit: append-only event log per case; FOIA reports generated automatically.
• Procurement and vendor management: architecture-of-record role across three implementation vendors; reference architecture acted as the contract surface.

Results

• 9 of 14 legacy systems decommissioned within 18 months; remaining 5 on phased path.
• Median cross-department case time dropped from 47 days to 12 days.
• Lost-case rate dropped from 9% to under 0.5%.
• AI routing classifier reached 91% staff acceptance rate after 6 months of feedback loop.
• Audit-trail completeness verified by national audit office.
• Cost of FOIA response per case fell by approximately 80% (reported by client compliance team).

Technology stack

.NET 10 (C#), PostgreSQL, Temporal for workflow, Redis, Azure Service Bus + Event Grid, RAG over case history, OpenAI / Claude for routing classification, side-by-side translation pipeline with deterministic fallback for legal text.