KuihDex

Inspiration

What it doesMost AI vision applications are highly functional but culturally sterile—they tell you "this is a green cake." We built KuihDex to prove that generative AI can possess hyper-local cultural DNA while directly driving physical commerce.

KuihDex is a mobile-first web application that allows users to snap a photo of any traditional Malaysian kuih. Under the hood, we pass the image to the Gemini 1.5 Pro Vision API using a strictly formatted JSON schema and a custom system prompt. Instead of a robotic textbook description, the AI embodies a traditional Malaysian "Makcik" (Auntie)—identifying the food, extracting the core ingredients (like Pandan or Gula Melaka), and delivering a highly opinionated, culturally accurate "roast" in authentic Manglish.

But identification is only half the battle. To bridge the digital-to-physical gap, we engineered an action layer. Once the Makcik identifies the kuih, the app dynamically generates a search query and pipes it into the Google Maps Places API. With a single tap on the Cari Dekat Sini button, users are instantly deep-linked into their native Maps app, routing them to the top three highest-rated local street vendors selling that exact item.

How we built it

Brain: Google AI Studio and Gemini 1.5 Pro API for multimodal image processing and structured JSON generation.

Muscle: Google Maps Places API for dynamic local business querying.

Body: A responsive, single-page React frontend built rapidly using Antigravity.

Hosting: Containerized via Docker and deployed live on Google Cloud Run (asia-southeast1) for zero-latency execution.