Qlue

Conversational AI with Actual Taste

Comment

Project Story

What Inspired Me

Qlue was born out of a personal frustration with how generic and passive most AI systems feel. They spit out answers, but they don’t really understand you. While working with Qloo’s cultural intelligence API, I had a lightbulb moment — what if instead of treating user preferences as static, I built an agent that actively investigates and discovers your taste patterns in real-time?

I kept coming back to one question: What if AI could truly understand someone by exploring their cultural DNA instead of just matching them to a fixed category?

What I Built

Qlue: The AI Agent That Actually Gets You

Qlue is my attempt to reimagine what personalized AI can feel like. It’s an intelligent agent that doesn’t just store your preferences — it investigates them, builds a live taste profile across domains, and feeds that intelligence into a video agent that speaks to you with real context and personality.

It doesn’t just react. It discovers.

The Core Innovation: Agent-Powered Taste Discovery

The real innovation was treating taste profiling as a living, evolving investigation. Qlue uses Qloo’s cultural intelligence API as its toolkit to:

  • Investigate what you like with targeted API calls

  • Discover hidden connections across domains (music → books → food → lifestyle)

  • Build psychological and cultural profiles in real time

  • Adapt its questioning strategy as it learns more

  • Stream insights to power conversations with the video agent

How I Built It

Technical Architecture & Challenges

The first major hurdle was wrangling the Qloo API. It’s powerful, but not simple to use directly. So I built a custom SDK from their OpenAPI spec, which became the core toolset for the agent.

Key Challenges I Tackled:

  1. Cross-Domain Signal Weighting
    Initially, the agent pulled in noisy results when jumping domains. I fixed this by implementing dynamic weighting to preserve relevance in context (e.g., music influencing food, but not overwhelming it).

  2. Pattern-Matched Profiling
    I realized users respond better to smart suggestions. Instead of throwing everything at them, I designed a profiling system where inputs in one domain intelligently influence recommendations in others:

- Podcasts → Books

- TV shows → Podcasts

- Books → Films

- Films → Books

  1. Agent Architecture Evolution
    I started with a fully agentic design using LLM tool calls, but eventually adopted a hybrid architecture — this gave me better control over flow, reliability, and context retention.

  2. Real-Time Streaming
    To keep users engaged, I implemented Server-Sent Events (SSE) so users could watch their profile evolve live as they interacted. It made the experience feel alive.

The Intelligence Pipeline:

User Input → Agent Processing → Qloo API Tools → 
Insight Discovery → Profile Enhancement → Video Agent

What I Learned

Technical Takeaways:

  • Tags are deceptively powerful — they unlock contextual richness.

  • Signal weighting is crucial when working across domains.

  • Hybrid agentic designs work better than pure autonomy.

  • Real-time feedback builds stronger user trust and engagement.

Product Takeaways:

  • Users don’t just want personalization — they want to feel understood.

  • The journey of uncovering your own taste is just as rewarding as the results.

  • Cross-domain patterns often reflect deeper psychology.

  • AI agents backed by real insights create far more authentic interactions.

The Impact

For Users:

  • Real Conversations: Agents that get you and speak your language

  • Taste Discovery: Unearth content and ideas you didn’t even know you’d love

  • Self-Awareness: Learn about yourself through the lens of culture

  • Growth Over Time: The more you interact, the deeper Qlue understands

For Businesses:

  • Marketing: Auto-generated rich consumer personas

  • E-commerce: Smart product pitches that feel human

  • Media Platforms: Taste-based discovery that actually surprises users

  • Dating & Lifestyle Apps: Compatibility beyond surface-level traits

The Future

I see Qlue evolving into a suite of specialized agents one for music, one for film, one for lifestyle — all collaborating to paint a multidimensional picture of the user. I’m also working toward turning Qlue into an intelligence layer that other platforms can plug into.

Built With

Share this project:

Updates