24HRS Messe-Lotse — Executive Summary

Date: 2026-06-10 | Status: Planned | Team: AI-only execution

The Product

A multi-user web application for managing trade fair booth projects. Replaces the current single-page prototype (index.html) with a production system backed by Pimcore, featuring AI-powered document processing, resource planning, and 3D booth generation. Scale: ~300 jobs/year, peaks at 113 jobs per fair.

Delivery Strategy: 6 Vertical Slices

Each slice delivers a complete, independently shippable user capability. No slice waits on a later slice.

 ┌────────────┐    ┌────────────┐    ┌────────────┐
 │  Slice 1   │───▶│  Slice 2   │───▶│  Slice 3   │   ← MVP: full trade fair
 │    Messe   │    │    Jobs    │    │  Workflow  │       management system
 └────────────┘    └────────────┘    └────────────┘
                                                        
 ┌────────────┐    ┌────────────┐    ┌────────────┐
 │  Slice 4   │    │  Slice 5   │    │  Slice 6   │   ← Intelligence &
 │ AI Skills  │    │ Resources  │    │ 3D Booth   │       automation layers
 └────────────┘    └────────────┘    └────────────┘

Slice	Capability Delivered	Value
1. Messe	Login, create venues, customers, and trade fairs; dashboard with fair list	Proves entire stack works; first visible UI
2. Jobs	Create booth projects linked to fairs/customers; master data editing; file upload via Pimcore DAM	Core entity relationships operational; replaces spreadsheet tracking
3. Workflow	All 8 tabs, 8-state workflow (Angebot→Abgeschlossen), Gotenberg PDF export, calendar view	Complete replacement of the index.html prototype
4. AI Skills	12 AI skills process uploaded documents — extract contacts, deadlines, budgets, materials; generate checklist items; detect brand consistency	Eliminates manual data entry from briefing documents
5. Resources	Worker, vehicle, equipment, accommodation pools with demand/assignment gap analysis; travel management	Replaces ad-hoc team/vehicle lists with structured resource planning
6. 3D Booth	One-click 3D booth model generation via SketchUp + Aluvision; PNG preview + BOM returned to job	Automated design pipeline from job data to production model

Architecture at a Glance

┌──────────────┐     ┌─────────────────┐     ┌──────────────┐
│  React SPA   │────▶│  Pimcore REST   │────▶│   MySQL DB   │
│  (TypeScript)│     │  API (OAuth2)   │     │              │
└──────────────┘     └────────┬────────┘     └──────────────┘
                              │
                    ┌─────────┼─────────┐
                    │                   │
              ┌─────▼─────┐    ┌───────▼──────┐
              │  Gotenberg │    │ Symfony      │
              │  (PDF)     │    │ Workflow     │
              └───────────┘    └──────────────┘

4-entity data model: FairLocation → Fair → Job ← Customer
4 user roles: Admin, Editor, Viewer, Customer
File storage: Pimcore DAM (not localStorage base64)
AI skills: Symfony Messenger queue with parallel workers

Key Decisions

Decision	Rationale
Vertical slices over sequential phases	AI requires small, verifiable increments; each slice independently shippable
Admin workflow override (`force=true`)	Real-world fair prep is messy; guards alone insufficient
Async 3D generation via message queue	SketchUp cannot handle sync HTTP from web controllers
Tiered document model with graceful fallback	Customer/fair/job-scoped docs, but zero-config for customers without brandbooks
LLM cost investigation before Slice 4	10,000+ calls/year needs cost modeling before commitment

Resource Pool (Slice 5) — Justification

The modeling of 11 resource entities is driven by real scale: CFC2026 alone has 113 booths across 113 customers at one fair, requiring coordinated worker (Barista, Rigger, Driver, etc.), vehicle, equipment, accommodation, and travel planning. A simple text field on the job entity would not support fair-level gap analysis.

Risks & Open Items

Risk	Status	Mitigation
LLM operational cost may exceed budget	⚠️ Investigation required	Defer non-critical skills if cost prohibitive
SketchUp must be running for 3D generation	⚠️ Accepted	Health monitoring + alerting; Slice 6 is optional nice-to-have
AI-only code quality	⚠️ Managed	Every slice ships with Playwright E2E tests; quality gate = test pass
Workflow guard rigidity vs real-world entropy	🔬 Investigation	Admin force transition implemented (Slice 3); guard strategy under review

Timeline

AI-only execution. Slices 1-3 form the MVP and can be verified independently. Slices 4-6 add intelligence and automation on top of the stable core. No external team dependencies; no developer hiring required.

Related: architecture-vertical-slices-1.0.md (full implementation plan, 368 lines, 100+ tasks)

5.8 KiB Raw Blame History