<<<<<<< Updated upstream ======= >>>>>>> Stashed changes
IMMOBILIEN.CHGroup
IMMOBILIEN.CH GROUP · Technology

Under the hood.

Five views of the platform that powers Immobilien.ch and Liegenschaften.ch — designed not as a catalogue but as evidence. The Architecture Matrix shows how we distribute responsibility across five pillars and three technical layers. The Fundament Circle makes visible that our Swiss foundation — cloud, FADP, FINMA-awareness — was built before the surface, not after. The Data Paths, the AI Building Blocks and the Tech Decisions show how a request travels, which models we control ourselves, and why, in the reboot after 33 years, we chose against the easier path.

Prototype A · Architecture Matrix
IMMOBILIEN.CH GROUP · Architecture

This is how Immobilien.ch is built. Five pillars, three layers.

Immobilien.ch (consumer marketplace) and Liegenschaften.ch (institutional deal platform) run on one shared architecture with five pillars: our consumer marketplace, our broker CRM, our property-management suite, our investor platform and the shared Data & AI Core. Every pillar stands on three technical layers — presentation, business logic, data & AI. Click any cell to see the concrete stack, the data model and the responsibility behind that one component.

This matrix is not a generic reference architecture — it is the actual construction we chose for the reboot of Immobilien.ch after 33 years. Every cell is a real technical responsibility, not a marketing tile.

Marketplace
Broker CRM
Property Mgmt
Investor Platform
Data & AI Core
Presentation
Search & Filter
REACT / TS
AI-first search field with semantic understanding — query is turned into a vector, matched against property embeddings. Frontend: React, TypeScript, Tailwind.
CRM Cockpit
REACT / TS
Broker workspace with kanban pipeline, deal timeline, client communication. Realtime updates via WebSocket, optimistic UI updates.
Enterprise Dashboard
REACT / TS
Family-office cockpit: KPI overview, ESG compliance, cash-flow analytics. Charts via Recharts, tables via custom grid.
Deal Room
REACT / TS
Off-market deal flow with audit trail, document vault, approval chains for regulated investors.
Explainable UI
REACT / TS
"Why this?" panels next to every AI suggestion — confidence score, feature importance, alternatives.
Business Logic
Ranking Engine
NODE.JS / GO
Multi-factor scoring: relevance × freshness × location match × user profile. Response < 100ms.
Pipeline API
NODE.JS
REST + GraphQL for every CRM object. Rate limiting, tenant isolation, audit logging.
Portfolio Engine
GO
Portfolio consolidation, cash-flow simulation, ESG scoring — for family offices and institutional investors.
Workflow Engine
GO / TEMPORAL
Long-running workflows (due diligence, closing) with state machine, event sourcing.
Inference Gateway
PYTHON / GO
Central API gateway for every AI model. Model versioning, A/B testing, caching.
Data / AI
Property Graph
POSTGRESQL
PostgreSQL 16 + pgvector for semantic search, PostGIS for geo queries.
Event Store
POSTGRES + KAFKA
All CRM-Aktionen als Events. Time-Travel-Queries, Audit-Compliance, Replay.
Time-Series DB
TIMESCALEDB
Metrics over time: rental income, NOI, cash flow. Continuous aggregates.
Document Vault
S3 / SWISS CLOUD
Encrypted document storage on Swiss soil. Content-addressable, versioned.
Valuation Models
PYTHON / PYTORCH
Swiss micro-market models: XGBoost für tabular, Transformer für sequential features. On-premise trained.
Prototype B · Fundament Circle
IMMOBILIEN.CH GROUP · Foundation

Built from the inside out. Immobilien.ch begins at the Swiss foundation.

The three rings show the order in which we assemble Immobilien.ch and Liegenschaften.ch — from the deepest layer to the visible surface. At the core: our Swiss foundation (Swiss cloud, FADP-compliant, FINMA-aware). Above it, the core services shared by both brands: authentication, data layer, AI inference, API gateway. On the outside, the four product surfaces: Marketplace (Immobilien.ch), broker CRM, property management, investor platform (Liegenschaften.ch). Click a segment to load the details.

Most new real-estate platforms build the reverse way: pretty UI first, compliance and data responsibility come later. During the reboot we deliberately chose against that. What you see as a user of Immobilien.ch sits on a foundation that was built first — not last.

Swiss Core
Foundation
Foundation
Swiss cloud infrastructure
The entire platform runs on Swiss cloud infrastructure — data centres with Swiss data-protection certification, FADP-compliant, FINMA-aware. No US operator, no EU backup, no grey zones.

Data residency: Switzerland · Encryption: at rest + in transit · Compliance: ISO 27001, FADP, FINMA-ready
Prototype C · Data Paths
IMMOBILIEN.CH GROUP · Data Paths

From input to answer. How Immobilien.ch handles your data.

Every interaction on Immobilien.ch or Liegenschaften.ch — a search query, a deal update, an investor request — passes through several clearly named layers. We show the three flows that actually happen on our platform: B2C (consumer search on Immobilien.ch), B2B (broker workflow in the CRM), B2I (investor mandate in Liegenschaften.ch). For each scenario you see which component sees data at which point — from user input through AI inference and the Property Graph to the answer back in the UI.

When you type "4.5-room apartment on Lake Zurich under CHF 2M" into Immobilien.ch, your data never leaves the Swiss perimeter — the semantic search runs on our infrastructure, the ranking uses our own models, and the answer returns in under 200 ms. Switch between the three flow scenarios to see how that is secured architecturally.

User Input FRONTEND AI Inference DATA + AI Property Graph POSTGRESQL Result UI
B2C · CONSUMER SEARCH: User types "4.5-room apartment on Lake Zurich under CHF 2M" → Semantic Query is turned into an embedding → vector-similarity search on the Property Graph → ranked results with confidence score. End-to-end < 200ms.
Prototype D · AI Building Blocks
IMMOBILIEN.CH GROUP · AI Building Blocks

24 building blocks, four layers. Everything that powers Immobilien.ch.

"AI-native" is not a marketing claim at Immobilien.ch — it is a nameable list. 24 concrete technical building blocks working underneath our platform, cleanly sorted by layer: Frontend (the UI components you see on Immobilien.ch), Backend (the services brokers and investors use), AI (the models we trained ourselves), Infrastructure (the Swiss foundation below). Click a tile to open the technical description.

Our view: anyone claiming an "AI-native platform" should be able to list the individual AI building blocks. Filter the grid by "AI" to see which concrete models and systems sit behind the answers from Immobilien.ch — no black box, no "AI magic" tile.

FRONTEND 6 building blocks · What the user sees
AI-First Search
Frontend
Realtime Sync
Frontend
Explainable AI
Frontend
Offline-First
Frontend
Design System
Frontend
A11y First
Frontend
BACKEND 6 building blocks · What brokers & investors use
API Gateway
Backend
Event Store
Backend
Workflow Engine
Backend
Document Vault
Backend
Pipeline API
Backend
Portfolio Engine
Backend
AI 6 building blocks · What we trained ourselves
Valuation Model
AI
Semantic Search
AI
Risk Scoring
AI
Deal Match
AI
Anomaly Detection
AI
Copilot
AI
INFRASTRUCTURE 6 building blocks · The Swiss foundation
Swiss Cloud
Infra
Observability
Infra
Zero-Trust
Infra
Backup & DR
Infra
CI/CD
Infra
Compliance
Infra
Prototype E · Tech Decisions
IMMOBILIEN.CH GROUP · Decisions

Six decisions of the reboot. Each with its counter-argument.

The reboot of Immobilien.ch after 33 years is not a continuation — it is a series of deliberate new technology decisions. We have documented every major decision, including the options we did not take: TypeScript over Python. PostgreSQL over MongoDB. Our own AI models over the OpenAI API. Swiss cloud over AWS Frankfurt. Zero-Trust over the perimeter model. GitOps over manual deployment. For each decision you see the counter-argument (the easier path), the pro-argument (harder, but the right one for Immobilien.ch) and the final choice.

"Responsible" does not mean, for us, that we never make mistakes — it means that every one of our foundational technology decisions is documented, traceable and open to challenge. Click a milestone — you read the same argument that was made internally during the reboot.

Q4 2025 Language choice · TypeScript
Counter-argument
Python for the backend would have made AI integration easier and put the team into model training faster.
Pro-argument
End-to-end TypeScript enables type sharing between frontend and backend, prevents whole classes of runtime errors and saves onboarding time for full-stack engineers.
Decision: TypeScript for frontend + backend. Python only for the AI training pipeline, in isolation.
Q4 2025 Database · PostgreSQL
Counter-argument
MongoDB would be more flexible for schema-less property data — many real-estate portals use NoSQL.
Pro-argument
PostgreSQL 16 with pgvector + PostGIS covers property search, geo queries AND semantic search in a single engine. ACID guarantees for financial transactions. One ops model instead of two.
Decision: PostgreSQL as the sole primary database. No second data model.
Q1 2026 AI strategy · Proprietary models
Counter-argument
The OpenAI/Anthropic API would work out of the box — no training effort, no GPU investment.
Pro-argument
Swiss micro-markets (Rapperswil vs. Herrliberg vs. Stäfa) have such specific price dynamics that generic models are systematically off the mark. Own data (33 years of domain history) + own models = defensibility.
Decision: Proprietary valuation models trained on Swiss soil. API models only for the language layer.
Q1 2026 Cloud choice · Swiss cloud
Counter-argument
AWS Frankfurt would be cheaper, offer more services and faster feature iteration.
Pro-argument
FADP compliance, FINMA awareness, "Swiss soil" positioning — for enterprise clients (Family Offices, institutional investors) this is a decision criterion, not a nice-to-have.
Decision: Swiss cloud provider with ISO 27001 + FADP certification. No cross-border backups.
Q1 2026 Security model · Zero-Trust
Counter-argument
The classic perimeter model (VPN + firewall) is well established and every developer knows it.
Pro-argument
Zero-Trust re-verifies every request: no implicit trust, tenant isolation at the database layer, MFA by default. Auditors now expect this.
Decision: Zero-Trust from day one. WebAuthn/Passkeys for users, mTLS between services.
Q2 2026 Deployment · GitOps + progressive rollout
Counter-argument
Classic CI/CD with manual approval is simpler to implement.
Pro-argument
Progressive rollouts (1% → 10% → 100%) with automatic rollback on error-rate spikes prevent full outages. GitOps makes every deployment reproducible and auditable.
Decision: GitOps via ArgoCD. Feature flags for gradual rollouts. Automated rollback.
Prototype F · Live Property Database
IMMOBILIEN.CH GROUP · Semantic Query Console

Natural-language query. AI-scored results. Live.

The Semantic Query Console demonstrates how IMMOBILIEN.CH will search: the user phrases the request in natural language, an AI parser breaks it into structured search parameters — region, room count, price range, features. Five rotary dials, six feature LEDs and a free-text field work in parallel; the scoring model weights every property along five dimensions and returns not merely hits, but explainable confidence values. Map, heatmap and waterfall visualize in real time how the match distributes across 50 curated Swiss properties. What works here as an interactive demo becomes the productive consumer platform with live market data in build phase 2027.

IMM.CH // SEMANTIC QUERY CONSOLE · v.0.4 · CH-North1 · 50 objects indexed --:--:-- CET
01 · Query Composer
query ›
awaiting input · tokens will appear here
Parsed AST
{ "region": null, "rooms": null, "price_max": null, "features": [] }
02 · Filter Rack
03 · Signature Dashboard
Query Confidence
Semantic clarity
Regional Heatmap
Low High
Top-Match Waterfall
Region
0%
Rooms
0%
Price
0%
Area
0%
Features
0%
Composite score0%
04 · Match Explorer
Regional Distribution
10
Cantons
50
Matches
--
Avg. Score
Live Matches · 50 objects
Semantic Tokens0
·
Confidence0%
·
Response-- ms
·
Objects50
·
Top Match--%
AI PARSER · REGEX v3 · OK · CH-NORTH1 · 142 ms · INDEX FRESHNESS · < 2h · SCORING MODEL · V.0.4 · QUERY LATENCY P95 · 187 ms · ACTIVE SESSIONS · 1 · AI PARSER · REGEX v3 · OK · CH-NORTH1 · 142 ms · INDEX FRESHNESS · < 2h · SCORING MODEL · V.0.4 · QUERY LATENCY P95 · 187 ms · ACTIVE SESSIONS · 1 ·
Prototype G · Compliance Console
IMMOBILIEN.CH GROUP · FINMA · FADP · ISO 27001

Compliance is not a slide bullet. It is a live system.

The Compliance Console shows in real time how Immobilien.ch and Liegenschaften.ch operate the four bearing regulatory frameworks: FADP (revised Federal Act on Data Protection), FINMA-awareness (FIDLEG Art. 4 for qualified investors), ISO 27001 (Information Security Management) and Zero-Trust network architecture. Each domain carries a compliance score broken down into four sub-criteria. A live audit log shows the last 30 data accesses classified by sensitivity. The data-residency map makes visible which Swiss datacenter processes which data type — no data leaves the Swiss perimeter. Click a domain card to see relevant log entries and affected datacenters highlighted — or a datacenter pin on the map for uptime, certifications and data categories in detail. Combines with the sensitivity filters; a click on »Reset filter« restores the overall view.

The platform itself is not subject to FINMA licensing; it provides the infrastructure through which qualified investors and brokers can structurally meet their own obligations under FIDLEG Art. 4, AMLA and FADP — with a complete audit log.

IMM.CH // COMPLIANCE CONSOLE · v.0.4 · CH-North1 · 4 domains monitored ALL SYSTEMS · OK --:--:-- CET
Filter active · Domain FADP Log entries, datacenters and runbook show only relevant items
01 · Regulatory Domains
02 · Data Movement Monitor
Live Audit Trail 30 events
Swiss Data Perimeter
Cross-border transfers · 0 All data within · CH
03 · Operational Readiness
Compliance Officer
Q2 2026
Legal mandate active · FINMA-experienced
ISO 27001 Audit
Q2 2027
Certification · prepared in build phase
Data Residency
100% CH
5 datacenters · 0 cross-border transfers
Composite Score--%
·
Events / hr--
·
Restricted Accesses--
·
Datacenters5
·
Uptime · 90d99.98%
Prototype H · Investor Dashboard
LIEGENSCHAFTEN.CH · Portfolio Simulator

Set investor criteria. Portfolio builds itself live.

The Investor Dashboard demonstrates how qualified investors will later assemble their portfolio on Liegenschaften.ch: the investor defines the investment thesis via controls — ticket range, target cap rate, time horizon, risk appetite, regions and asset classes. A deal universe of 36 curated Swiss properties is scored in real time and visualized in the scatterplot as a point cloud. The best-matching deals are combined into a sample portfolio and broken down economically in the waterfall chart — from gross rental income via operating costs to net yield. Every control movement triggers instant recomputation; no real deals, no live backend — a frontend simulation of the institutional investor view.

LIE.CH // INVESTOR DASHBOARD · v.0.4 · CH-SOUTH1 · 36 deals indexed --:--:-- CET
01 · Investment Thesis Composer
Ticket Range 200k – 10M CHF
Target Cap-Rate ≥ 3.5%
Time Horizon5 years
Regionsalle
Risk Appetite
Balanced
Asset Classesalle
02 · Deal Universe
Cap-Rate × Ticket-Size Dot size = Deal Quality 36 / 36 in thesis
Cap-Rate (%) Ticket Size (CHF)
03 · Portfolio Preview
Top-Match Deals 8 of 36
Match Deal Region Cap-Rate NOI / year Ticket
Portfolio yield
Weighted Cap-Rate
--%
weighted by ticket size
Total Investment
CHF --
sum of all portfolio deals
Diversification
--%
Regions × Asset classes
Avg. Deal Quality
--%
score from internal rating
04 · Deal-Economics Waterfall
From gross rent to net yield weighted portfolio average
--
--
--
--
--
Gross rent
− Op. costs
= NOI
− Taxes
= Net
In Thesis--
·
Portfolio Size8
·
Total CHF--
·
IRR--%
·
Response-- ms
Prototype I · Federated Broker Network
IMMOBILIEN.CH GROUP · Broker-to-Broker Matching

Match off-market. Without exposing data.

The Federated Broker Network shows how broker agencies can match offers and demands with each other without any raw data leaving the agency boundary. Six brokers from different cantons exchange encrypted queries through a central matching layer; each match card in the live feed shows paired offer and demand with match score, NDA status and canton origin. Three privacy tiles explain the zero-knowledge principle step by step: raw data stays in the agency, only hashes travel the network, details are exchanged only after mutual NDA handshake — with FINMA audit log. A 30-day timeline with sparkline and canton-pair matrix shows network activity. Click a broker to see their outbound queries; click a privacy tile to expand the ZKP principle; click a match card for encryption details.

IMM.CH // FEDERATED BROKER NETWORK · v.0.4 · Zero-Knowledge Layer · 6 broker nodes · 247 matches/30d --:--:-- CET
01 · Broker Network · Encrypted Query Layer
02 · Live Match Feed · Cross-Canton
8 active matches · Offer ↔ Demand Live
03 · Privacy Guarantees · Zero-Knowledge Principle
Zero-Knowledge Match
Broker A only sees the match score, no details of Broker B's demand. The match is cryptographically proven without revealing the underlying data.
01Broker A hashes their offer (price, area, canton) with SHA-256
02Broker B hashes their demand identically
03Matching layer compares hashes homomorphically
04Only the match score is returned to both brokers
Zero Data Transfer
Raw data never leaves the broker agency. Only pseudonymised hashes travel the network. The matching layer stores no customer, price or address data.
·Customer data: encrypted on-premises in the agency
·Offer hashes: temporary in matching layer (24h TTL)
·Match results: bilateral, encrypted, no broadcast
·No central data store — no attack surface
On-Demand Reveal
After mutual NDA handshake, the brokers exchange details bilaterally. All reveal steps are documented in the FINMA audit log — provable compliance.
01Both brokers accept NDA (digitally signed)
02Encrypted channel established (TLS 1.3 + E2E)
03Details exchanged · audit log written
0410-year retention per FINMA FIDLEG Art. 4
04 · Network Activity · 30-Day Trend
Matches per day · 30-day sparkline
Matches last 30 days
247
NDA handshake rate
89%
Ø Match time
2.3h
Cross-canton share
68%
Top canton pairs · 30-day activity
Fully FINMA-compliant · FADP-compliant · Based on federated learning principles as in the European Data Space Standard. No competitor in Switzerland has this architecture.
Broker Nodes6
·
Active Matches8
·
Data Transfer0 bytes / broker boundary
·
EncryptionSHA-256 · TLS 1.3
MAIN PAGE