nestAI

An AI agent that searches the biggest accommodation platforms and recommends the best options based on your context, not just price and stars.

Context

nestAI

nestAI is an AI agent that automates accommodation search across major booking platforms, reasoning about the specific context of your trip rather than stopping at numeric filters. Users describe their situation in plain language, "I have a three-year-old", "quiet area away from nightlife", "no car", "I want to visit the Colosseum", and the agent translates those preferences into motivated recommendations, extracting candidates via browser automation, evaluating them with an AI model, and verifying availability for the exact requested dates. Results stream live in the web interface and arrive as a Discord notification with a direct booking link.

What we did on it...

Vision

Project conception

The idea came from the difficulty that long-term travellers and digital nomads face when searching for accommodation: traditional platforms filter by price and stars, but they don't understand the trip. What was needed was an agent capable of reasoning about real-world contexts, a family travelling with young children, a traveller without a car, someone wanting to stay close to the places they want to visit, and returning a motivated answer, not just a list.

Structure

Agent architecture

The agent runs in three distinct phases: scraping the platform via browser automation with filters baked directly into the URL, qualitative reasoning over a set of candidates, and availability verification for the exact dates. The system prompt is assembled dynamically, only the relevant reasoning blocks are activated based on preferences detected in the user's input. The architecture is designed to scale across sources.

Interface

Interface development

The web interface is built on FastAPI with real-time streaming via Server-Sent Events, keeping the user updated phase by phase. Results are displayed as cards showing price per night, rating, location, the agent's reasoning, and availability status. The interface supports Italian and English and is complemented by an interactive CLI for terminal use.

Launch

Testing and release

The agent was tested on real destinations with diverse travel profiles: families with children, tight budgets, car-free travellers. The full cycle of scraping, reasoning, and verification completes in under 60 seconds on average. Results are also delivered as formatted embeds on Discord.