Qloofy

Inspiration

We wanted to reimagine what a digital concierge could be — not just one that knows your location, but one that understands your cultural soul. Qloo’s Taste AI offered a rare opportunity to blend cultural intelligence with conversational AI. We saw the chance to bring that to life in a delightful, Telegram-native experience.

Whether someone is exploring a new city, looking for a date night idea, or simply craving a vibe-matching plan, we wanted to build an assistant that understands tastes, not just preferences — and does so privately.

What it does

Qloofy (TasteBot) is a Telegram-native AI concierge that crafts personalized cultural plans based on your mood, tastes, and location. It integrates:

• 🎨 Taste Profiling: Understands your unique cultural fingerprint using Qloo’s API
  
• 🤖 Multi-LLM Reasoning: Orchestrates GPT, Claude, and Gemini for flexible responses
  
• 🌍 Geo-Based Plans: Integrates Foursquare & Geoapify to suggest local spots
  
• 🔍 Real-Time Enrichment: Uses Tavily, Firecrawl, and SERP APIs to fetch live context
  
• 💳 Credit System: Manages paid usage with Stripe
  
• 📱 Seamless Chat UX: Fully embedded in Telegram with webhook support
  

It delivers suggestions like:

• “Take me on a chill Saturday adventure in Lisbon”
• “Where should I eat if I like Tarantino films and soul music?”
• “What’s a cozy bookstore that fits my vibe?”

All of this — with no personal data collected.

How we built it

We built a modular backend using Node.js, NestJS, and PostgreSQL, with:

• 🧠 Qloo Taste AI API for semantic cultural graphs
  
• 💬 LLM orchestration via OpenAI, Groq, and Gemini
  
• 📍 Location logic via Geoapify and Foursquare
  
• 🔎 Search enrichment via Tavily, Firecrawl, and custom SERP wrappers
  
• 💳 Stripe integration for credits and monetization
  
• 📲 Telegram webhook bot with native UX
  
• ⚙️ Prisma ORM and clean code separation by feature module
  

Challenges we ran into

• 🔀 Balancing LLM latency and model variability during orchestration
  
• 🧩 Harmonizing structured (Qloo) and unstructured (SERP, Tavily) data into one plan
  
• 🔒 Maintaining privacy while building rich user context
  
• 🧠 Designing fallbacks and fail-safes inside Telegram flows
  
• 🌐 Matching place recommendations with real-time web enrichment in meaningful ways
  

Accomplishments that we're proud of

• 🎯 Created a fully working Telegram cultural assistant in a few weeks
  
• 🌐 Live-tested real-world plans using all APIs simultaneously
  
• 🔀 Built clean orchestration across 3 LLM providers
  
• 🧠 Designed a privacy-first taste profiling flow
  
• 🚀 Integrated 8+ APIs in a coherent product
  
• 🛠️ Scalable backend with modular architecture and production-readiness
  

What we learned

• Qloo’s taste graph is incredibly powerful when combined with contextual LLMs
  
• Real-time search (Tavily + Firecrawl) makes AI feel smarter
  
• Prompting styles must adapt per model (Claude vs GPT vs Gemini)
  
• Telegram bots can be beautifully complex and lightweight products
  
• Productizing taste-based AI requires clear UX and fallback thinking
  

What's next for Qloofy

• 🌍 Expand to WhatsApp and Web interfaces
  
• 🗣️ Add multi-language support and localization
  
• 🧩 Let users fine-tune or share plans like "mood playlists"
  
• 📊 Build taste dashboards powered by Qloo embeddings
  
• 🧑‍💼 Launch white-labeled bot for tour guides & influencers
  
• 🧠 Integrate LangChain memory and Redis session caching
  
• 🛒 Explore Shopify-style use for culturally relevant eCommerce
  

Built With

• built-with:-typescript
• firecrawl
• foursquare
• gemini
• geoapify
• groq
• jinai
• langchain
• nestjs
• node.js
• openai
• postgresql
• prisma
• qloo-taste-ai-api
• redis
• serp
• serper
• stripe
• tavily
• telegram-bot-api