Embedding-based matching and voice-to-listing for a community trading app. iOS, Android, and a web admin panel, all from one FlutterFlow build.
A mobile-first community platform for trading goods and services locally, without the friction of classified ads or Facebook Marketplace chaos. Two AI capabilities sit at the core: semantic matching and natural-language listing creation. iOS, Android, and a web-based admin panel, all shipped from one build.
The challenge wasn't building a marketplace. It was making two things feel effortless: finding the right match between what someone has and what someone else wants, and getting people to actually list in the first place. Keyword search misses too much. Manual listing forms kill supply on day one.
Two AI features had to anchor the product. First, an embedding-based matching engine that converts every listing into a semantic vector, so "vintage lamp" matches "retro home decor" without sharing a single word. Second, an AI-powered listing flow where users describe what they have in natural language, by text or voice, and the app returns a structured draft ready to publish.
Scoping those two flows first meant the architecture supported them from day one, instead of treating them as bolt-ons.
The stack: FlutterFlow for cross-platform mobile, Supabase for the database, and Node.js for custom backend logic where FlutterFlow alone wasn't enough.
The matching engine runs on OpenAI text embeddings. When a listing is created, its title and description are converted into a vector and stored in Supabase using pgvector. New listings are compared against existing ones by vector proximity, surfacing semantically similar results regardless of exact wording. Firebase push notifications fire match alerts in real time.
The AI listing input takes voice or text, passes it to a language model, and returns structured drafts with pre-filled fields. Users review and publish in seconds.
The rest of the platform, listing management, in-app chat, offer and request workflows, reputation ratings, and a moderation-ready admin dashboard, was built and shipped in the same release. The platform launched on time with both AI features live and fully integrated.
30-minute discovery call. Scope, timeline, and fixed quote out the other side. No strings attached.