Housing Platform, Finland

Context

A Finnish housing company wanted to move beyond static listing pages. Buyers were dropping off because finding the right property meant navigating dozens of filters. The brief was to let people describe what they want in plain language and get relevant matches back, with the ability to book viewings directly from the conversation.

Constraint

Finnish language support was non-negotiable, and the AI had to work with a property database that updated in near real-time as listings came and went. The conversation had to feel natural in both Finnish and English without separate codebases, and every recommendation had to trace back to a live listing.

What I built

A conversational AI layer on top of an existing property platform. The system takes natural-language queries ('three-bedroom apartment near Kallio under 350k'), retrieves matching listings from the live database, presents them with images and details, and lets users book viewings or ask follow-up questions.

Decisions

• Used Azure AI for language understanding with custom fine-tuning for Finnish property terminology.
• Built retrieval against the live listing database rather than a periodic snapshot so results are always current.
• Kept the conversation stateful across sessions so returning users pick up where they left off.
• Designed the UI as a panel alongside the existing site rather than replacing it, so users can switch between browsing and conversation.
• Added structured fallback: when the AI cannot parse a query, it offers filter-based alternatives rather than a generic error.

Outcome

Property enquiry conversion increased after launch. The conversational interface handles Finnish and English queries, and viewing bookings through the AI channel grew steadily through the first quarter.

Stack

• Azure AI
• TypeScript
• React
• Next.js
• Node.js
• REST APIs
• PostgreSQL