How Interfaze serves 50 million requests a month on Vercel
By Steven Van ·
Interfaze runs its API, code sandbox and provider fallback routing on Vercel, using Fluid compute, Vercel Sandbox and AI Gateway.
Interfaze runs its whole backend on Vercel, including the API that serves 50 million requests a month. Vercel describes the setup in a customer story, and Interfaze has also published its own post about its latest model.
Interfaze builds models for deterministic work that general-purpose LLMs handle poorly: document processing, OCR, object detection, web scraping, speech to text and classification. Its models return a bounding box and a confidence score with each result, so an answer such as an invoice total can be checked against where it appeared on the page.
Backend as serverless functions
Interfaze's frontend and backend share one codebase on Vercel. The backend handles every request to the API, whether a customer is calling the model or uploading a file, and it runs as serverless functions. When a request needs computation, the model hands the code to Vercel Sandbox, which runs it in isolation. Founder Yoeven Khemlani says no separate infrastructure had to be set up: "On Vercel, I create a folder and I have a backend."
Fluid compute for heavy jobs
Scaling with traffic was Khemlani's top requirement, and Fluid compute scales on its own. Interfaze also runs heavier work there, including web scraping, because Vercel's memory and timeout limits are high enough for it.
Fallback routing with AI Gateway
Interfaze puts multiple model providers behind one fallback layer using AI Gateway. This routing keeps data preprocessing running when a third-party model provider goes down.
