Building Enterprise AI for a €50M German Construction Firm

Building Enterprise AI for a €50M German Construction Firm

Building Enterprise AI for a €50M German Construction Firm

How I Delivered 12 Microservices in 3 Months

How I Delivered 12 Microservices in 3 Months

How I Delivered 12 Microservices in 3 Months

THE CHALLENGE Simply BAU, a traditional German construction and property management SMB, was drowning in manual processes. With operations split across multiple business units—construction, property management, and client services—they had no centralized data, no automation, and critical business decisions were being made from scattered Excel files and email chains. The complexity went deeper: they needed to operate in both German and English, comply with strict European data regulations, and integrate with legacy German business software that most international developers had never heard of. But the real challenge? They wanted enterprise-grade automation on startup costs, delivered by a single consultant in 3 months.

Company:

Simply Bau

My Role:

Solution Architect (AI / Infra / Automation)

Year:

2025

Techstack

Postgres | Azure AI Search | NextJS | GPT-4o | Redis

THE SOLUTION


AI-First Microservices Architecture with German Business Process Intelligence

Instead of building a monolithic solution, I designed a microservices ecosystem where each service solved a specific business problem—but AI connected everything.

🧠 1. Document Intelligence Platform (Azure Cognitive Search + Postgres)

The Foundation: 23,027 Business Documents Connected to 500,000+ Business Records

I built an integrated AI-powered intelligence platform that connects document search with live business data:

  • Multi-language AI processing (German/English) with proper linguistic analysis

  • Real-time document classification using Azure Cognitive Services

  • Business entity extraction (projects, customers, contractors, compliance docs)

  • 4 specialized search indexes optimized for different document types


Business Intelligence Layer (PostgreSQL + 53 MCP Tools):

  • Live connection to 50,000+ business records across customers, projects, invoices

  • Smart questionnaire processing with automatic customer/project creation

  • K-### readable customer IDs (K-336, K-337...) with deduplication

  • Intelligent project code generation (City + Street + Number = BH42)


Integrated Intelligence:

When a document mentions "Sanitärinstallation for Project BH42," the AI instantly knows:

  • Project BH42 = Berlin Hauptstraße 42 (from PostgreSQL project code)

  • Customer K-336 owns this property (from customer database)

  • €15,000 budget allocated for plumbing (from project calculations)

  • Last invoice dated March 2024 (from invoice records)

  • Related compliance documents in folder SH/ (from document classification)


This wasn't just search or just data—it was connected business intelligence where documents and data created a complete business context.


🔧 2. Business Process Orchestration (Core Microservices)

9 Services, One Intelligent System

I built 12 specialized microservices that operated as a unified AI-driven platform:

  • Core Orchestrator: Cross-service workflow management with event-driven architecture

  • Microsoft Graph Integration: Full M365 automation (18+ mailboxes monitored)

  • Excel Service: 33 MCP tools for Projektkalkulation workflows

  • Questionnaire Service: Next.js + MCP combined deployment for lead qualification

  • Business Intelligence: PostgreSQL analytics with German business metrics

  • Folder Automation: OneDrive structure automation following German business standards


Each service included production monitoring, health checks, and automatic recovery procedures.


🇩🇪 3. German Business Email Automation - Technical Architecture

Built a production-ready email automation service with hybrid webhook/polling architecture and comprehensive MCP management:

Hybrid Processing Architecture:

  • Real-time Microsoft Graph Webhooks: Immediate processing with bulletproof token management

  • Fallback Polling System: 10-second intervals to catch missed emails with deduplication

  • Circuit Breaker Pattern: Graceful failure recovery with exponential backoff

  • German Name Semantic Processing: Proper parsing with titles/prefixes, service provider exclusion

AI-Powered Email Classification Pipeline:


```javascript

// 3-Step Processing Pipeline

  1. Entity Extraction: German business terminology understanding

  2. Customer Classification: Fuzzy matching with blacklist filtering

  3. Path Deduction: Specialist routing (Joe: Bathroom/Heating, Greg: Painting/Plastering)

```

Orchestrator-Triggered Email Generation:


```javascript

// Questionnaire → Email Workflow

webhookData.event_type === 'questionnaire_completed'

→ Rule Matching (database-driven configuration)

→ AI Gap Analysis (missing documents identification)

→ German Template Processing (Handlebars + project data)

→ Specialist Assignment + Outlook Draft Creation

```

MCP Management Ecosystem (50+ Tools in 5 Categories):


Email Generation (12 tools): `create_email_generation_rule`, `preview_template_with_data`, `update_specialist_assignment`

Specialist Management (8 tools): `get_specialist_workload`, `update_specialist_availability`, `route_project_to_specialist`

Analysis & Prompts (10 tools): `configure_gap_analysis_prompt`, `update_german_business_prompts`, `test_analysis_accuracy`

Monitoring (12 tools): `get_email_generation_stats`, `audit_rule_performance`, `track_processing_metrics`

Configuration (8 tools): `update_database_rule_config`, `manage_webhook_endpoints`, `configure_attachment_filtering`


German Business Email Intelligence:


Generated emails like: "Sehr geehrter Herr Müller, basierend auf Ihrer Anfrage für die Badezimmersanierung haben wir eine erste Projektanalyse durchgeführt. Für Projekt BH42 fehlen noch: Grundriss-Pläne, Sanitärinstallation-Genehmigung..."


  • But not templated—each email was dynamically generated by:

  • Project Data Analysis**: Questionnaire responses + PostgreSQL project records

  • Gap Analysis AI**: OneDrive folder inspection for missing documents

  • Specialist Intelligence**: Workload balancing + availability checking

  • German Business Standards**: Professional terminology + compliance formatting


Technical Implementation Details:


Database-Driven Rules: PostgreSQL-backed rule configuration with JSON-RPC 2.0 compliance

Bulletproof Token Management: Proactive renewal 1 hour before expiration with circuit breaker

Attachment Intelligence: Pre-filtering to remove logos/signatures before AI analysis

Audit Trail: Comprehensive logging with performance analytics and error tracking

Outlook Integration: Direct draft creation in specialist's "AI Generated Emails" folder


This created a self-managing German business communication system where every customer interaction triggered contextual, intelligent responses that understood project requirements, identified gaps, and routed to appropriate specialists—all managed through sophisticated MCP tooling.

💡 4. Claude Desktop Integration (150+ Tools)

AI-Powered Business Operations Interface

We create an admin dashboard but we felt it better to wrap all of the microservices and create 150+ MCP tools that allowed the Simply BAU team to manage their entire operation through Claude or any other AI client.

  • Email generation and management tools

  • Excel automation for project calculations

  • Document search and analysis capabilities

  • Business intelligence reporting and analytics

  • System monitoring and configuration management


This meant they could manage a complex enterprise platform using natural language: "Generate a quote for the Munich bathroom renovation project" or "Find all compliance documents for the Berlin construction site."


📊 THE TECHNICAL ACHIEVEMENTS


Real Production Metrics That Matter

✅ 23,027 documents indexed and AI-searchable across German/English

✅ 12 microservices in production with zero-downtime capability

✅ €543-763/month operational costs (vs €10,000+ enterprise systems)

✅ 150+ Claude Desktop tools for business operations

✅ Multi-language AI processing with German business terminology

✅ Production-ready containerized deployment with disaster recovery


But the real achievement? I delivered enterprise-grade AI automation to a traditional German SMB that had never worked with cloud services before.


🎯 THE BUSINESS IMPACT

From Manual Processes to AI-Driven Operations


Within 3 months, Simply BAU transformed from a traditional paper-based operation to an AI-powered business:

• Complete email automation with German language processing

• Intelligent document management replacing manual filing systems

• AI-powered project cost calculation and quote generation

• Real-time business intelligence replacing Excel-based reporting

• Automated OneDrive organization following German business standards

The most impressive result: they could now handle 3x more client inquiries with the same team size, while maintaining German business compliance standards.


🔍 THE CONSULTING REALITY

Managing Complex Stakeholders Under Pressure


This wasn't just a technical challenge—it was a masterclass in client management:

• Cross-cultural communication (German business formality vs. startup agility)

• Framework agreement navigation with compliance requirements

• Scope management when clients wanted 300% more than contracted

• Technical education for non-technical stakeholders

• Timeline management with dependencies across multiple business units


Key Learning: Traditional European SMBs need AI solutions that respect their existing business processes, not Silicon Valley disruption that breaks everything.


Next Project