Kelify is Spain’s first AI-powered real estate marketplace, designed to modernize a fragmented housing market. Home seekers have long been frustrated by disjointed listings and rigid search filters on legacy portals, while property owners struggle with overwhelming inquiry volumes and time-consuming tenant screening.

CEO Javier Cuevas and his co-founder launched Kelify.com in 2025 to connect home seekers and private property owners through a seamless, automated experience. “We wanted to streamline the housing market by intelligently matching properties with qualified seekers, providing a transparent, frictionless experience,” says Javier.

The team achieves this by aggregating millions of property listings across Spain. Home seekers browse these listings via Keli, Kelify’s AI search assistant, which allows for conversational or photo-based searches. An AI Voice Agent also assists property owners by answering calls, fielding questions, screening tenants based on specific criteria, and automatically booking viewings.

The challenge: Molding raw data into intelligent matching

The Kelify team faced three main challenges that traditional code could not solve.

1. Unstructured data normalization: “We aggregate millions of real estate listings from diverse sources. The raw data is messy, often with inconsistent descriptions or hidden attributes, such as ‘no pets’ buried in a paragraph,” explains Javier. “We had to find a way to standardize over five million listings into a clean, structured database without manual entry.”

2. Bridging language and database structures: To allow users to discover homes based on “vibes” and lifestyle needs, the search tool needed to translate nuanced requests into precise database queries—something legacy portals couldn’t do. As Javier notes, “A checkbox can filter for ‘terrace’, but it can't find ‘a sunny spot for morning coffee’."

3. Low-latency voice capabilities: The team needed a 24/7 receptionist for property owners. “Our voice agent needed to understand intent, qualify leads by asking about income or job type, and respond naturally in under a second. Traditional chatbots were too slow and robotic.”

Recognizing the promise of generative AI’s reasoning capabilities and speed, the team turned to Gemini.

The solution: Vibe-coding breakthroughs with the Gemini API

The Kelify team first experimented with other AI models, but ultimately settled on Gemini models like Gemini 3 Flash and Gemini 3 Pro. “We found that Gemini 3 Flash offered a superior balance of speed, cost, and a larger context window compared to OpenAI’s GPT-3.5 and GPT-4,” says Javier. The team began implementing the Gemini API as the core engine of their platform in stages.

Laying the foundation

The first stage involved integrating Gemini 3 Flash into the backend to solve "hard data" problems. By using multimodal capabilities, the Kelify team extracted structured data from raw text and photos. Gemini Flash identified desirable amenities—like "natural light" or "modern kitchen"—to automatically populate filters and standardize listings at scale.

The conversational search breakthrough

Next, the team unlocked a natural language search experience. “We vibe-coded a prototype of our conversational AI search using Google AI Studio,” Javier explains. “We fed our GraphQL schema directly into the model and realized Gemini could instantly map natural language intent to our complex database structure.”

In just two weeks, the team moved from an AI Studio prototype to a production version of the Keli Search Assistant.

“We use Gemini 3 Flash to build our natural language search. It translates complex requests, like ‘find a three-bedroom flat in a hipster neighborhood,’ into precise database queries.”

The AI voice agent

Kelify harnessed the low latency of Gemini Flash to build its most critical tool for property owners. “Gemini Flash handles the logic for inbound calls, dynamically generating responses based on property details and the caller’s answers,” says Javier. “It ensures the conversation flows naturally while adhering to the owner’s qualification criteria.”

Throughout development, the team utilized Gemini 2.5 Pro and Gemini 3 Pro as intelligent pair programmers, which multiplied engineering velocity and helped debug complex agentic workflows.

The impact: Prototyping in hours, not weeks

Since integrating Gemini models, the Kelify team has seen transformative results across marketing, engagement, and development:

• 35% increase in average session duration

• 22% higher user retention among users interacting with Keli

• 100,000 monthly organic clicks in 6 months

The team also used Gemini models to execute their content strategy. As Javier explains, “We transform raw data into unique, hyper-local market guides for every city and neighborhood on our platform. This helps us capture high-intent search traffic, growing from under 2,000 to over 100,000 monthly organic clicks in just 6 months with €0 in ad spend.”

For Javier, the engineering impact was equally profound. “Using Gemini Pro as a coding assistant helped us ship complex agentic workflows at startup speed. Keli was prototyped in just a few hours and deployed to production in two weeks—something that would have taken a specialized team four to six months with traditional development methods.”

What’s next: Automating the entire rental journey

Javier and his team are now evolving Kelify from a discovery platform to a fully transactional ecosystem. “We want to move beyond just finding a home to automating the application, vetting, and agreement phases entirely within the platform,” says Javier. Expansion into other European countries is also on the horizon.

Javier is passionate about the role AI can play in supercharging startups. “AI has changed the economics of our industry, allowing a lean founding team to compete directly with massive incumbents.”

Reflecting on his experience, Javier offers simple advice for fellow founders: “Don't over-engineer the start. If we had to start over, we’d prototype in AI Studio first, validate the magic, and only then write the production code.”

“The barrier to entry has never been lower. You don't need to be a machine learning expert to leverage these tools. Using Gemini 3 Pro as a coding partner and Gemini Flash as your product engine allows you to move at a speed that feels unfair to competitors who aren't using them. Stop waiting and start vibe-coding.”