Relationship Intelligence Platform in 4 Weeks
Nordic Professional Services Firm · March 15, 2026
The Engagement
A Nordic professional services firm needed a bespoke intelligence platform to map and navigate a 20-year relationship network across IT, insurance, finance, and banking. Traditional CRMs forced relationships into linear pipelines — they needed a graph-based system where connections, companies, pursuits, and contracts form a living web of interconnected intelligence. The scope was ambitious: a full-stack production application with graph database modelling, five embedded AI agents, SSO authentication, GDPR-compliant data handling, and Azure cloud deployment — all from a blank repository.
What Was Delivered
The factory delivered the complete platform in 4 weeks with 6 parallel AI workers:
- Graph-powered relationship engine using PostgreSQL with Apache AGE, modelling contacts, companies, pursuits, and contracts as interconnected nodes and edges
- 5 embedded AI agents — Smart Scribe (automated data structuring from raw input), Quick Dossier (contact enrichment), Pathfinder (relationship path discovery), Recommendation Narrator (strategic recommendations), War Room (pursuit planning)
- 20+ application pages including capture workflows, graph visualisation, action tracking, contract management, and an operational dashboard
- 40+ backend services covering audit logging, deduplication, health scoring, evidence trails, and provenance tracking
- Microsoft Entra SSO with role-based access control
- Full Azure deployment on Container Apps with infrastructure-as-code (Bicep), Key Vault, managed PostgreSQL, and CI/CD pipelines
- GDPR compliance by design — three-layer data architecture separating immutable artifacts, append-only observations, and rebuildable derived views
Results
- 446 issues closed across the full delivery
- 483 pull requests merged with a median merge time of 10 minutes
- 81% of issues resolved within 24 hours
- 6 parallel AI workers delivering simultaneously
- 0.6% escalation rate — only 12 of 1,979 work sessions required human intervention
- 4 weeks from empty repository to production deployment