Conversational AI · shipped · 2026 · Client: Contract — UAE Wedding & Events Client (via Innovacio Tech)
A bilingual AI consultant that plans UAE weddings by matching couples to real vendors across 20 categories conversationally.
Planning a wedding or large event in the UAE means coordinating 20+ vendor categories — venues, photographers, kosha, catering and more — often bilingually and with cultural requirements like women-only events. Vendor discovery and matching are slow, fragmented and manual.
A bilingual conversational backend built on a LangGraph orchestrator: a supervisor router sends each message to a general agent that collects event basics or to one of 20 category agents that ask vendor-specific questions and match real vendors from a structured dataset. Vendor facts are delivered from data through typed SSE events — never generated by the model — for zero hallucination.
A single prompt handling venues, photographers, kosha, catering and 16 more categories collapsed under its own context — questions bled between categories and the model re-asked things it had already been told.
We split it into a LangGraph supervisor that routes each message to either a general agent collecting event basics, or one of 20 category agents that ask vendor-specific questions. Cross-category memory means guest count or a women-only requirement is captured once and never asked again.
Vendor prices, contacts and ratings are never generated. They are served from structured data through typed SSE events, so hallucinated vendors are impossible by construction rather than by prompt instruction.
A supervisor routes each turn to the right specialist. Vendor facts flow from the database through typed events — the model writes the language, never the data.
OpenAI GPT-4o, LangGraph, Zod, Node.js, Express, Server-Sent Events, React 18, Vite, Tailwind CSS, Railway, File-based JSON store