Host Agent — Where Different Tastes Find Common Ground

Inspiration

We've all been there: planning a movie night, dinner, or gathering with friends, only to discover that everyone wants something different.

One person loves jazz, another prefers electronic music. Someone wants Italian food, while someone else prefers Japanese cuisine. Traditional recommendation engines usually optimize for one person, and generic AI assistants often suggest whatever is popular without understanding the cultural connections between people's interests.

We wanted to explore a different question:

What if an AI agent could understand multiple people's cultural preferences and help them discover experiences together?

That's why we built Host Agent, powered by Qloo's Taste AI.

What it does

Host Agent is a group-oriented cultural discovery and planning application that helps people explore recommendations based on their combined preferences.

Users can create a group of 2–4 guests and provide taste signals for each person, such as favorite artists, films, music genres, and cuisines.

Host Agent processes these preferences and generates recommendations across four categories:

• Music: Artists and musical inspiration for the gathering.
• Dining: Food and restaurant-related discovery.
• Film: Movie recommendations for the group.
• Vibe: Lifestyle and brand inspiration to help shape the gathering's atmosphere.

Instead of relying entirely on a language model's general knowledge, Host Agent uses Qloo's cultural taste graph to discover relationships between interests across different domains.

Qloo-powered recommendations vs. a generic baseline

A central feature of Host Agent is its comparison between Qloo-grounded recommendations and a generic baseline.

The baseline produces fixed, general suggestions without calling Qloo. The Qloo-powered workflow resolves the group's interests into real cultural entities and tags, then queries Qloo for recommendations informed by those signals.

The interface distinguishes LIVE, SAMPLE, and BASELINE recommendations so users can understand where each result comes from.

How we built it

We built Host Agent using Next.js, React, TypeScript, and Qloo's Taste AI API.

The application includes a server-side recommendation workflow that:

1. Collects and validates preferences from multiple guests.
2. Resolves those preferences into Qloo entities and cultural tags.
3. Builds group-level queries using the resolved taste signals.
4. Retrieves recommendations across music, dining, film, and lifestyle categories.
5. Normalizes, validates, deduplicates, and ranks the results.
6. Presents recommendations with transparent source labels and available supporting evidence.

We also implemented bounded API concurrency, timeout handling, request-budget safeguards for testing, and partial fallback behavior to keep the experience understandable when upstream requests fail.

Challenges we faced

1. Working with real cultural data

Integrating the Qloo API required understanding how entity searches, tag lookups, and insight queries work together.

We had to handle different response structures and distinguish between entities, genres, and cuisine tags rather than treating every preference as a simple text search.

2. Latency across multiple recommendation categories

Our first full live planner test completed, but dining and film requests timed out.

We improved request scheduling and increased the insights timeout while keeping concurrency and the overall execution deadline bounded.

In a subsequent end-to-end test, all four recommendation categories returned LIVE Qloo results in 13.015 seconds, with 12 successful API requests and no fallback.

3. Trustworthy recommendation explanations

We wanted to avoid presenting guesses as evidence.

Host Agent does not invent affinity percentages or claim that a particular guest likes a recommendation unless supporting Qloo results are available.

When individual compatibility hasn't been checked, the interface communicates that limitation rather than treating missing evidence as dislike.

4. Making the experience usable

We designed a responsive interface that makes it easy to enter multiple guests' preferences, explore recommendations, and compare cultural intelligence with a generic baseline.

What we learned

Building Host Agent taught us that cultural intelligence is more than generating plausible recommendations.

It requires structured preference resolution, reliable API integration, careful interpretation of evidence, and honest communication about uncertainty.

We also learned that combining group preferences is not the same as proving unanimous agreement. Our current system offers group-informed recommendations, while stronger individual compatibility and consensus modeling remain opportunities for improvement.

What's next

We envision Host Agent becoming a more complete AI-powered social planning assistant.

Future improvements include:

• Stronger guest-level compatibility and consensus ranking.
• Location-aware dining and activity discovery.
• Dietary, accessibility, and budget preferences.
• More cultural categories and personalized event planning.
• Better explanations of why a recommendation fits the group.

Our goal is simple: make planning together easier, more culturally aware, and more personal.

Built With

• ai
• api
• css
• next.js
• node.js
• qloo
• react
• rest
• tailwind
• taste
• typescript