Embed AI as a practical screening assistant users can rely on, not an opaque system.

Remove the conversion gap before "Submit inquiry," where users stalled because metrics were unclear and they doubted the choice.

Make the key parameters work as a one-glance answer for compare-and-decide.

Keep filters concise to avoid overwhelming users with options or accidentally exclude relevant listings.

Tet users collect, return, compare, prepare for discussion, and convert to inquiry.

The platform must allow changes to selection logic, metrics, and fields without a rebuild.

We built the frontend as a Next.js application split into UI components and service modules — map, listing, filters, shortlist, property page, request form. We designed the search state as controlled and shareable: filter parameters, sort order, map viewport, and selected property all sync into the URL so users can navigate back, share a view, or refresh without losing progress.

We chose Directus as the managed backend core for collections, roles, access rights, editorial and operational workflows, and the admin panel. On top of Directus, we added a Node.js API layer for dedicated search endpoints, metric aggregation, shortlist handling, request creation, and integrations (notifications, maps, AI). This split responsibilities cleanly: Directus handles data and administration; Node.js handles product logic and integration flows.

We designed the data model around metrics as first-class structured fields, enabling fast filter/sort/compare/aggregate. We also built in flexibility — adding new metrics and selection rules should not break cards, filters, or the property page.

We integrated AI as a separate service that receives a base candidate list and returns a ranked result with short relevance explanations. The principle: AI stays a secondary assistant that narrows options faster, while the decision remains transparent through visible metrics and explanations.

We built the Maps API integration so the map and list operate in lockstep — selecting a marker highlights its card, hovering or clicking a card highlights the marker, and viewport changes preserving the active selection. This is critical for commercial real estate, where location is a primary decision factor.

We implemented the request as a full record with context: selected property, shortlist, key metrics, active filters, and user comment captured at the moment of submission. This reduces context loss between the interface and the next step in sales operations (broker contact, qualification, tour). Notifications are connected through a dedicated provider and adapter so each request reaches the right role immediately.

live AI-assisted marketplace with a managed admin layer and a stable technical foundation, ready to put in front of partners and early clients within ~3 months from kickoff

key metrics surfaced on cards and property pages, supporting ~2× faster shortlist creation vs. the founder's original spreadsheet workflow

ranking and short explanations help users identify relevant options faster and understand the recommendation logic

saved properties become a working selection process: users return, compare, and convert into an inquiry rather than abandoning

every request preserves selection context (property, shortlist, filters, key metrics), so the broker team receives concrete inquiries and can move users to the next step faster

Directus lets the operations team update content and reference data without engineering involvement; the modular architecture absorbs new metrics, selection rules, and integrations without rewrites