Colombia's first AI-native real estate portal
Introduction
Seamless AI designed the full product — brand, design system, conversation architecture, and UI — for a consumer real estate portal serving Colombian and international buyers at once.
A Texas-based founder relocating to Floridablanca, Santander arrived with brand instincts, a locked name, and a strategic framework. Seamless AI built everything else, from the first brand token to the final Claude Design prompt. No Colombian portal had attempted an AI layer like it.
Problem
The warmth problem. Most conversational AI is a form wearing a chatbot costume. The advisor had to feel like a trusted local friend, not a lead-capture tool that asks for budget first.
The dual-market problem. The platform had to serve Colombian and international buyers in the same session — COP and USD on one card, Spanish and English in one interface, estrato badges that matter in Bucaramanga and mean nothing in Dallas.
The no-infrastructure problem. No Colombian portal had an AI layer and no design system existed for this market. Every Colombia-specific component had to be built from scratch.
Research
Competitive landscape. The market leaders carried no AI and no dedicated experience for international or diaspora buyers.
Regional relationships. The Lonja de Propiedad Raíz de Santander and its 90 affiliated brokerages were identified as the Year 1 anchor.
Market geography. The Bucaramanga metro, and the Cañaveral corridor specifically, were mapped as the proof-of-concept market.
The warmth of a top realtor relationship could be delivered digitally, at scale, to a market that had never experienced it. The platform would compete on trust, not inventory.
Serving two markets at once was an architecture problem, not a translation problem.
The durable advantage would be proprietary data and broker trust — AI features are the entry point, not the moat.
Design process
Studied how Colombian buyers and international buyers each experience a market that has historically felt cold and transactional.
A consumer portal where an AI advisor holds the relationship, budget is never the first question, and the platform owns the conversation across sessions.
A master brand with four products on one color system, a design system built on Airbnb's architecture with Tinto tokens applied 1:1, and a conversational layer as a first-class navigation element.
Twelve brand slides built in Figma via MCP, plus grid and property-detail mockups and the advisor-panel prompt package in Claude Design.
Solution
Brand and identity system
Master brand Tinto with four product suffixes, mission and values, three elevator pitches, bilingual copy rules, and a forbidden-words list.
Design system as seven token files
Structured so any AI design tool ingests the full set and produces on-brand output — DM Mono locked to all data surfaces, DM Serif to editorial moments only.
Ten Colombia-specific components
AVM card, dual COP/USD display, estrato badges, VIS/VIP badges, predial chip, valorización chip, ES/EN toggle, m²/sq ft display, the advisor panel surface, and the mobile in-thread conversation card.
The advisor and its conversation spec
Four buyer archetypes, a complete bilingual conversation tree, the trust sequence, dual-pane grid logic, Property Detail Mode, and the observation engine.
Testing & iteration
The product is pre-launch, so this case study anchors on provable scope. The outcome numbers below are held until the platform is live in the Bucaramanga metro.
Account creation rate, advisor engagement rate, or saved-search conversion — needs live launch data.
The six-month pre-arrival plan is built around three proof assets: a market intelligence dashboard, a working AVM demo, and an international demand engine.
Reflection
Trust is designed at the structural level — the trust sequence and the contact model, not just the copy.
A dual-market product is an information architecture challenge before it is a translation one.
Building an entire product in pre-arrival mode is possible when brand, design system, and conversation logic are locked before a single screen is drawn.