PropVR AI → Craft

The problem

Architects, interior designers, developers and property marketers were already using AI, but the workflow was fragmented. A general prompt box could produce a useful image. It could not guide someone through concept development, material edits, staging, panoramic media, animation and delivery on its own.

The product question was not which model to expose. It was how AI should fit the way AEC teams already move work through a project.

The foundation

I built an internal MVP with 22 specialist tools organised across five workflow studios. Each tool owned a clear job, input contract and output. The product routed tasks to suitable models rather than treating one model as the whole product.

The initial stack used Next.js 15, Tailwind CSS 4, AI SDK integrations, fal.ai, OpenAI and Gemini-family image and vision models. Production persistence, real billing and account infrastructure were deliberately outside the MVP scope. State was client-side. The job was to establish the workflow structure, tool boundaries, model routing and user experience for a team to extend.

Why specialist tools came first

A blank prompt box gives the model freedom and leaves the user responsible for the rest. Many professional tasks need constrained inputs, useful defaults, predictable output formats and a clear next action.

The studio structure made the lifecycle visible. A user could choose a stage, run a focused capability, inspect the result and carry it forward. The design intent was to make prompt construction an implementation detail rather than a prerequisite for using the product.

The trade-offs

Choice	What it made possible	What it did not solve
Specialist tools	Clearer input and output contracts for each task	End-to-end coordination across a whole project
Model routing	Choose a capability suited to the job	A guarantee that every generated result is correct
Client-side state	A small, quickly extensible MVP scope	Durable projects, production accounts or real billing
Five workflow studios	A recognisable way into the tool set	Proof of production adoption or measured business impact

The handoff

The MVP was handed to PropVR’s team as a foundation to extend. My contribution was the initial product system and implementation. The wider team developed, productionized and expanded that foundation into the current platform.

Initial PropVR AI MVP: Pranshul Chandhok. Current Craft platform: PropVR Technology team.

The product lesson

A model can generate an output. A product has to decide what the output means, what must remain fixed, who approves it, and what happens next.

The MVP established a way to arrange specialist capabilities around AEC work. Its central lesson was to define the task, the input contract and the next decision before choosing the model.