Flux

Flux helps users find products through natural conversation. Ask about laptops, headphones, or any product - the assistant searches real inventory and displays results alongside the chat. The interface splits between conversation and live product cards, creating an interactive shopping experience. ## Technical Implementation ### Core Stack - **LLM**: Groq API with `llama-3.3-70b-versatile` for tool calling - **Speech**: Groq Whisper (`whisper-large-v3-turbo`) for voice transcription - **Vision**: Groq Vision (`llama-3.2-11b-vision-preview`) for image analysis - **Video Avatar**: Tavus API for real-time AI video conversations - **Search**: Tavily API for product discovery - **Framework**: Next.js 15 with App Router - **Styling**: Tailwind CSS v4 ### Key Features **Multi-Agent System** Three specialized agents handle different domains: - Marketplace agent for product searches - Travel agent for flight and hotel queries - Food agent for restaurant recommendations **Real-Time Product Search** When users mention products, the system: 1. Extracts search intent using the 70B model's tool calling 2. Queries Tavily for real product data 3. Displays results in the interactive feed **Context Extraction** All conversations are parsed to extract: - Dates, locations, prices - Product names, flight numbers - User preferences and requirements This structured data appears in the activity feed, providing visual confirmation of what the AI understood.