Pichet: from Symfony/eZplatform to AI Vision on property floor plans.

Pichet, French premium property developer. A catalog of hundreds of new-build programs, each with its 2D/3D floor plans, descriptive notes, units, and commercial status. Our engagement: move their web catalog from an aging eZplatform to a modern platform, then graft AI Vision today to automatically interpret floor plans.

Starting point (2018-2020)

Adrien D. embedded on the tech side, lead on the catalog back-end overhaul. Initial stack: PHP 7/8 + Symfony 4/5 + eZplatform for the CMS, Docker + Kubernetes + AWS for runtime. Progressive migration from a legacy monolith to a containerized architecture, deployable multiple times a day.

What was delivered

• CMS Symfony / eZplatform overhaul: 280 programs migrated in 9 months with no service downtime
• Automated asset production pipeline (plans, axonometric drawings, photos)
• Multi-AZ EKS Kubernetes cluster with cleanly versioned Helm charts
• Datadog / Prometheus observability, on-call alerting for commercial peaks
• Documentation and skills transfer to Pichet's internal team

The stack we ship today

On the same business (new-build catalog, buyer journey, conversion on 400-800 k EUR units), here's what we assemble today:

• Next.js 16 + React 19 on Vercel: Server Components by default, ISR per program, edge stock fetch
• Sanity as headless CMS: data model for programs / units / typologies / updates edited by marketing
• Claude Vision on 2D/3D floor plans: automatic extraction of living area, orientation, room layout, accessibility constraints
• pgvector + embeddings: semantic search 'bright 3-room apartment near a park' returns the right units
• Snowflake data pipeline to track buyer funnel through to notarial signing
• Hybrid Vercel (public front) + AWS (back office and internal services) with audited network boundary

Why it's hard

• Heterogeneous plans: 6 architecture firms, 6 graphical conventions, 0 standardization
• Sensitive commercial data: availability, reserve price, options under negotiation -> strict RBAC
• Long lifecycle: a program goes through 18 months of pre-marketing then 24-36 months of delivery
• Critical local SEO: majority of traffic arrives via 'new build [city]', clean sitemap and schema.org RealEstateAgent / Apartment required
• Multi-stakeholder coordination: sales, construction tech, marketing, legal, syndic

What this engagement taught Abbeal

Pichet was our first serious immersion in premium B2C real estate. We learned to code for plans that are never standardized, to respect developer commercial tempo (windows are tight), and to ship a platform that serves both marketing and after-sales. That's the experience we replay today with other developers and with social landlords industrializing their catalog.