Deterministic models, verifiable output
Interfaze builds models for tasks general-purpose LLMs handle poorly: document processing, OCR, object detection, web scraping, speech to text, and classification. The distinction matters because those tasks need results a human can check. An LLM asked for an invoice total returns a number with no traceable origin; Interfaze's models return the number alongside a bounding box locating it on the page and a confidence score. The output stays auditable without giving up prompt-driven interaction.
The API, the sandbox that executes model-generated code, and the provider routing layer all run on Vercel. That covers 50 million requests a month.
Backend and sandbox in one codebase
Frontend and backend both live on Vercel. Every API request — model calls and file uploads alike — runs as a serverless function in the same codebase as the UI. When a request requires computation, the model passes the code to Vercel Sandbox, which executes it in isolation. No separate infrastructure had to be provisioned.
“On other platforms, setting up a backend means networking, security, and function limits before you serve a single request. On Vercel, I create a folder and I have a backend.”
— Yoeven Khemlani CEO & Co-Founder @ Interfaze
Fluid compute absorbs the load
Scalability was the founder's top requirement, since the goal is serving the model to millions of users. Fluid compute scales with traffic without manual intervention. The demanding jobs run there as well: Vercel's memory and timeout limits are generous enough that Interfaze's Puppeteer-based web scraping works unchanged.
“We never expected to use Vercel for our backend. Fluid compute is a real Node.js server underneath, so even our heaviest workloads just work.”
Failing over between model providers
Interfaze sends large volumes of data cleaning through open source models hosted by third parties. A single provider becomes a single point of failure: rate limits and outages halt the cleaning unless traffic can move. Providers are also not equivalent — two can claim to host the same open source model and serve it at noticeably different quality.
AI Gateway addresses both problems. It fails requests over to the next provider automatically when one throttles or fails, and it lets Interfaze pick which provider serves a given model. That second capability lets the team run the same workload through each provider, compare results, and direct traffic to the better one.
“Two providers will claim they're hosting the same model. One's at four bit, one's at eight. With AI Gateway we can route between them and see who's actually doing better.”
Lifecycle email with durable state
The Workflow SDK runs Interfaze's lifecycle email system, where sends depend on product usage. The system tracks which API a customer called, at what volume, and how that compares with their first week. Retries happen automatically, so a single failed step doesn't break the sequence. Durable execution was the decisive feature: a workflow can pause a step and resume it months later with its state intact — something the team found other systems couldn't sustain over that horizon.
Deployments as version boundaries
Interfaze serves more than one model version concurrently, each at a distinct URL. On Vercel, every URL is its own deployment with its own logs, so an incoming error is already attributed to the version that produced it, and no separate logging vendor is involved. Those preview URLs double as a release channel: a new model version becomes a new branch whose preview link goes to selected customers for testing before merge. Building the Vercel side of a release takes about three minutes; hosting the weights on GPUs is the slower step.
“A new model version is a new branch. I send the preview URL to a few customers, they test it for a month, and we merge.”
Next: an open-weight model and a search gateway
Interfaze recently released Interfaze 1 Lite, an open-weight version of its model that runs on a single GPU and is served through the same API. A web search gateway is next — one API routing to whichever search provider a developer selects, mirroring what AI Gateway does for models — and it launches on the same Vercel backend. Interfaze's customers are engineering teams in legal, healthcare, and finance.
