2026 Case file
PropVR AI → Craft
Initial MVP delivered to PropVR’s team
I built the initial 22-tool AI MVP across five workflow studios. PropVR’s team used that product and technical foundation to develop Craft, a public creative platform for AEC workflows.
On this page
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.
So the product question 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.
A tool contract, reconstructed
Consider a material-editing task: start from a room image, identify the change, run the appropriate model and inspect the returned image. This is a conceptual reconstruction of the MVP’s tool structure, rather than a capture of its interface.
| Moment | Product decision |
|---|---|
| Choose the task | Enter a specialist tool through its workflow studio |
| Provide the input | Make the source image and requested operation clear |
| Run the tool | Route the operation to a suitable model |
| Inspect the result | Return an output the user can assess before choosing the next task |
I separated those responsibilities so the team could extend one tool without redefining the whole product, and so the task boundaries and model integrations were ready to hand over.
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
I handed the foundation to PropVR’s Technology team, who developed, productionized and expanded it into Craft. The homepage at the top of this page shows their later platform; the tool contract above describes the earlier MVP.
Initial PropVR AI MVP: Pranshul Chandhok. Current Craft platform: PropVR Technology team.
The product lesson
The MVP taught me to define the task, the input contract and the next decision before choosing a model. A model can generate an output; the product has to decide what that output means, what must remain fixed, who approves it and what happens next.
From an internal foundation to Craft
Craft’s tools index now lists 30+ production tools live across six studios: Concept, Design, Immersive, Animation, Refine, and Marketing. Direct tools can be used individually. Agentic 1.0 and Arc 1.0 are in early access, adding outcome planning and shared project context around those tools.
That is the PropVR Technology team’s platform today. What I built is the initial MVP in the first column of the table below.
- Initial MVP22 tools across five studiosPranshul Chandhok
- Team handoffProduct and technical foundationPropVR Technology team takes it forward
- Current Craft30+ production toolsLive across six studios
- Agentic 1.0 + Arc 1.0Early accessOutcome planning and project context
| Initial PropVR AI MVP | Current Craft platform |
|---|---|
| 22 tools across five workflow studios | 30+ production tools live across six studios |
| User selects a specialist tool | Direct tools plus agent-planned workflows |
| Client-side prototype state | Shared project context through Arc in early access |
| Proved workflow structure, tool boundaries, and model routing | Adds production operations, approval, consistency, and cost controls |
Availability checked: .
Go deeper