BUILD Use Cases / Overview
Keep the app simple.
Keep the app mainly read-only: BUILD is not designed for user entry or complex UI interactions.
Focus on data analysis or data visualization with client-specific calculation, logic and customized UI presentation.
Focus on these visualizations: Table, Plotly Chart, Markdown/simple html
BUILD Limitations
Read the documentation overview to understand the general architecture of an app.
Pickers capabilities are limited to what's implemented in the frontend codebase. No amount of vibe coding can alter or add new features to the pickers. Make sure to cross-reference what's possible in the doc.
Brainstorming
Figure out what to build before prompting.
Think through the data analysis, what are the use cases/goals and who are the target users e.g. (Engineer bottleneck analysis, live operator dashboard, historical executive KPI etc).
Find the data. Look at the data table, field explorer, dataviz to explore and find the data for the analysis.
Plan the sidebar pickers: Do you need a date picker or two? Are the asset/field hardcoded or selectable? Etc.
Decide on the visualizations and layout: an app can have multiple panels and visualizations.
Consider app page linking to create a click-through workflow e.g. one visualization to another.
Creation / Iteration
Start with full app creation. Prompt it with all the details from above: what picker to include, what the app does, what visualizations it should have. The more detail or concrete the better.
Vibe coding works best early and/or on smaller and more specific tasks.
Use restore revision history often and earlier when vibe coding fails.
Read the code.
As the app becomes more finalized, consider using non-vibe-coding chat or external AI tools to ask specific questions. Use the answer to manually update the code. This avoids the vibe-coding breaking major existing work.
Validation
Always validate the visualization output. Go to data table and dataviz to check and match the math and calculation.
Double check all date picker selection is handled.
Review
Ask chat what this app does.
Use revision history to return to a version / figure out what changed. Revision history can run/preview a previous version.
Read the code.
Migration
Clone app by downloading and uploading.
An app is just a sidebar json + a notebook json combined.
Apps can be moved from one environment to another (but be careful of client sensitive / specific code)
Advance
It's possible to store and read CSV files and ML models in the notebook. This enables analysis to use pre-trained models (like ML recommendation values, risk etc.) or pulls values from CSV files. The setup is a bit convoluted for now, so reach out to #factory-app-builder for help.
It's possible to run BUILD on an interval to do tasks like alerting, recurring reporting etc. (example like generate a weekly KPI csv report). This is also a bit difficult to set up, so reach out to #factory-app-builder for help.
BUILD is running legacy model (GPT-4-1, which is a year old) and before we migrate to newer models, it can be useful to download the app JSON files and use newer models to help generate new apps. The external AI won't have context on BUILD, so the JSON downloaded is all the context it will have. This approach is especially useful when you want to make another app that generally has the same structure (picker, asset to query etc.) as the downloaded app but want other types of calculation and visualization.