// Key takeaways notes when implementing Questions for Widgets (Fund Analysis Report) //
Questions work better using .json for text only
AI at times get carried away and skipping steps when finding and replacing HTML codes
AI risks hallucinating when codes break due to more than 2+ syntax errors, or when finding and replace codes are done, but in the wrong widgets sections even when no syntax errors.
Best way is forcing AI to find function, then provide exact find and replace html codes. Requires uploading index.html first as a guideline
AI doesn’t prefer more than 200 rows of code at a time, so best to attach index.html after every instance
Best practices for our workflow is using hard coded values first for each question, then have AI update calculations while using JSON values for guideline during testing
Finalize UX and CSS before testing calculations, then finalize layout and tweak as necessary when testing results
When updating requires several steps, best to save version after every update and have AI verify index.html after every 3 updates
Force AI to show remaining steps as a summarized checklist after every few updates before proceeding
AI hallucinates when starting its next Widget update regarding its processes, so best to save your workflow phases design as an image (or copy and paste), then revert back as reference
Do not let AI make recommendations and solely focus on following the initial process
Prior to API integration, we used API URL placeholder, then updated API functions including fetch, before replacing actual results for questions
Netlify allows for deployment without publishing saving AI credits when testing in preview
Google Apps Script allows for JSON to filter out null data
Other Widgets sections can still work if another section has a syntax error or being updated
If you plan on using this Model with your own data, you can replace with your own APIs and .json
