Generative UI Dashboards for BIM Models with json-render and APS Viewer
A practical look at generating viewer-first BIM dashboards from natural language prompts using Vercel json-render, AI SDK, and Autodesk Platform Services Viewer.
Frame Team
Generative UI dashboards are becoming one of the more interesting directions for BIM analytics. The idea is simple: instead of opening a dashboarding tool, choosing visuals, wiring filters, and building report pages manually, you ask a question in natural language and the application generates a useful dashboard interface from that prompt.
That does not mean replacing tools like Power BI or Tableau. Those platforms are still the right choice for governed reporting, multi-source data models, enterprise distribution, and long-term analytics pipelines. But there is a real gap between having a question about a model and getting an interactive visualization in front of the team.
This prototype explores that gap.
The Concept
The prototype starts with a chat interface and prompt suggestions. A user asks for a dashboard, and the application generates a viewer-first BIM dashboard for a fixed Autodesk showcase model.
The generated dashboard can include:
- Filters and slicers
- KPI cards
- Bar charts, pie charts, and line charts
- Data tables
- Element detail panels
- An embedded Autodesk Platform Services Viewer
- Viewer-linked chart and table interactions
The important part is that the generated dashboard is not just a static AI response. It is an interactive interface. Clicking a chart segment or table row can isolate related elements in the 3D viewer. Selecting an element in the viewer updates the detail panel with the related model properties.
Why This Matters For BIM Teams
BIM data is rich, but it is often difficult to explore quickly. A model can hold thousands of elements, categories, families, types, levels, materials, and custom properties. The data is useful, but only if the team can get to the right view of it.
Traditional dashboard tools are powerful, but they introduce setup work:
- choosing the right data source
- shaping the data model
- selecting visuals
- building filters
- configuring interactions
- publishing and sharing the report
That workflow is worth it for mature reporting. It is slower when the team just wants to ask, “What is in this model?” or “Show me wall areas by level and type.”
Generative dashboards are useful because they reduce the distance between the question and the first usable interface.
A Viewer-First Dashboard Shell
The prototype is intentionally viewer-first. Every supported dashboard keeps the Autodesk Viewer as a primary section of the layout because BIM analytics needs spatial context.
For example, a generated dashboard might include:
- a filter band for level, category, family, or material
- KPI cards for total elements, total area, or selected quantities
- a large 3D viewer panel
- a primary chart for the requested breakdown
- a secondary chart or comparison
- a table with grouped model data
- a property detail panel for the selected element
This structure makes the dashboard feel closer to a BIM coordination interface than a generic BI page.
Editing After Generation
The dashboard is not locked after the AI generates it. A user can continue editing the output by:
- adding supported visuals
- changing chart types
- reordering widgets
- removing sections that are not useful
- retargeting widgets to different model fields
- opening shared dashboards in a full-page editor
This is important because AI rarely gets the perfect dashboard on the first try. The better pattern is to let the model generate a strong starting point, then let the user shape it.
Handling Vague Prompts
Not every prompt should immediately become a dashboard. If the request is too vague, the prototype can suggest enriched options before generation.
For example, instead of trying to generate from “show me the model,” the app can offer more useful dashboard prompts such as:
- “Create a wall takeoff dashboard by level and type.”
- “Show model element counts by category and family.”
- “Build a material breakdown with a table of related elements.”
This keeps the interaction fast without forcing the user to understand dashboard design or model schema details upfront.
How The Prototype Works
The project is built with Next.js, Vercel AI SDK, and Vercel’s
json-render packages.
json-render describes itself as a generative UI framework: the model emits a
structured UI specification, and the application renders that specification
through a controlled component catalog.
In this prototype, the flow looks like this:
- The user sends a dashboard prompt.
- The app checks whether the prompt needs refinement.
- The agent queries normalized model data from the Autodesk showcase model.
- The AI model generates a structured dashboard UI spec.
- The app normalizes the spec into a viewer-first dashboard shell.
- The dashboard renders with charts, filters, tables, and APS Viewer.
- The user can edit and share the generated dashboard.
That controlled rendering step matters. The model is not generating arbitrary frontend code. It is choosing from supported components, chart types, filters, and layout regions that the app knows how to render safely.
Why Not Just Use Power BI?
Power BI is still the better tool for many use cases.
Use Power BI when the job requires:
- multiple data sources
- governed semantic models
- organizational reporting standards
- scheduled refresh
- role-based access
- enterprise distribution
- long-term dashboard maintenance
Generative UI dashboards are better for fast exploration:
- asking immediate questions about a model
- creating a temporary dashboard for a meeting
- giving non-analysts a way to inspect model data
- testing which visual breakdowns are useful
- moving from model data to interface without a full BI setup
The point is not to replace Power BI. The point is to create a faster path from “I have a question about this model” to “I can see and interact with the answer.”
Where This Could Fit In Frame
For Frame, this direction is especially interesting because our core workflow is already model-to-dashboard. We help teams connect BIM data to Power BI, Excel, and interactive 3D viewers. Generative UI adds another layer: the dashboard can start from a question instead of a predefined report template.
Potential use cases include:
- quick quantity takeoff exploration
- model health investigations
- room and asset summaries
- category and family breakdowns
- coordination dashboards
- prompt-driven viewer filtering
- temporary dashboards before a formal Power BI report is built
This could become a practical bridge between raw BIM data and governed BI reporting. A team could explore the model quickly with generative UI, then promote the useful patterns into a Power BI template or recurring Frame report.
Try The Prototype
The prototype is available here:
Try the generative BIM dashboard prototype
Reference project:
Related Frame posts: