@Wednesday our company has a hack day. Here’s the documentation log of what I’m building, using AI app building tools.

PHASE 1: GOALS & SCOPE

Problem

We make product or operational changes constantly but our FAQs and SOPs do not keep up with the pace of our changes.

Evidence

• We have various new currencies launched but FAQs still don't reflect those changes (sample link)
• Back and forth on asking about fees updates on Feb 2025 when updates were made in 2024

• Manual need to create & update FAQs -> takes up time when all this can be automated. Especially annoying when there is a BIG update (e.g. Thunder)

• Other companies not updating their documentation -> annoying

Proposed Solution

A way to track product updates or frequently asked questions from sources like support tickets and suggest regular updates to FAQs to keep documentation in sync.

MVP Scope & Implementation

• User can connect their current set of FAQ documents
• Input links on FAQ documentation to read from
• If time: Actually connect to Zendesk API
• User can import support tickets from last X days
• Connect to Zendesk API
• Import 7 days worth of tickets
• If time: Also allow the import of other documentation such as PRDs on Notion, tickets on Linear, or just plain text
• User will then see suggestions for 1) new FAQs that should be created; 2) edits to existing FAQs
• New: Show list of new FAQs and suggested topic titles. When you click into the suggestion, it also previews all the support tickets with links out related to that suggested topic.
• Edits: Show list of all existing FAQs and which ones have potential to be edited. When clicked, it shows the diff highlighting what the suggested edits are. It also previews all the support tickets with links out related to the edits to that FAQ.
• If time: Merge or cut: Show list of FAQs that can be cut out
• Inspiration for UI: x.com/marty_kausas/status/1879748003013009764
• User can now add these suggestions to their knowledge base
• Easy copy/paste LEL
• If time: Add text editor for easy edits + push to connected Knowledge Base

PHASE 2: INITIAL PROMPT

Prompt with Lovable.dev

Build an AI-powered support knowledge manager that helps users refine and expand their FAQs based on actual support tickets.

Sidebar Menu with the following:

1. Inputs (shows the "Inputs" section below)
2. Suggestions (which shows the "Suggestions" section dashboard below)

Inputs section

1. Connect Existing FAQ Documents
• Allow users to input links to their FAQ documentation. The system should scrape and extract content for analysis.
• (Stretch goal) Provide an option to connect directly to the Zendesk API to pull existing FAQ content.
2. Import Support Tickets from the Last X Days
• Default: Import the last 7 days of support tickets from Zendesk.
• (Stretch goal) Allow importing from additional sources like:
• PRDs from Notion
• Tickets from Linear
• Plain text files (drag & drop upload)
3. ChatGPT API Analysis
1. After this, we will call the ChatGPT API and ask it to
1. Look at the support tickets from the last 7 days and also look at the existing FAQ documents
2. Based on support tickets, it must analyze if a NEW article needs to be created that was asked in support tickets that doesn't already exist in the knowledge base and if there are suggested edits to the existing FAQs. The output will be shown as below.

Suggestions section

The main Suggestions dashboard then shows up after they completed the info in the Inputs tab, otherwise an empty state shows.

1. Generate FAQ Suggestions & Edits with 2 sub tabs
1. First Tab: "New FAQ"
1. This tab shows a table with column headers "Suggested Topics", "Related Tickets", "Create Article"
2. "Suggested Topics" shows the suggested title of the new Knowledge Base article that should be made
3. "Related Tickets" shows the # of tickets that contributes to the mention of that suggested topic, analyzed by AI using ChatGPT API
4. "Create Article" shows a button that when clicked, opens a large modal that allows the user to create an article based on that suggested topic. The modal then pops up and generates the contents of that topic. There is also a copy/paste button on this modal that, when clicked easily copy/pastes that article.
5. When the line item on the table is clicked, a side drawer opens up that shows a preview of relevant support tickets that mentions that particular topic, line by line, previewing the subject of that support ticket (from zendesk) and a link to go to that support ticket on Zendesk
2. Second Tab: "Edits to Existing FAQs"
1. This tab shows a table with column headers "Suggested Articles to Edit", "Related Tickets", "Update"
2. "Suggested Articles to Edit" shows the title of the existing Knowledge Base article
3. "Related Tickets" shows the # of tickets that contributes to the editing of that suggested topic analyzed by AI using ChatGPT API
4. When the line item on the table is clicked, a side drawer opens up that first shows the suggested edits to the article like a diff checker. After, it shows a preview of relevant support tickets that mentions that particular topic, line by line, previewing the subject of that support ticket (from zendesk) and a link to go to that support ticket on Zendesk
5. "Update Article" shows a button that when clicked, opens a large modal that allows the user the text to the updated article. The modal then pops up and generates the contents of that topic. There is also a copy/paste button on this modal that, when clicked easily copy/pastes that article.

UI/UX Considerations

• Simple, intuitive dashboard
• A clean, card-based layout in light mode
• Any primary buttons or any are the color #276D6A
• Support tickets should be linked inline as cards for easy reference.
• For article creation or edits, use a side-by-side diff view to highlight suggested changes on the modal

This app should prioritize automation and ease of use, reducing the manual effort needed to maintain a high-quality FAQ repository.

PHASE 3: REFINING WITH AI

Learnings

• If I want a backend, I should connect the backend off the bat. Lovable makes it easy with Supabase. I can also do authentication!
• Knowledge manager allows me to add universal prompts. Good for code templates or UI guidelines.
• Configure everything auth related upfront & make sure the database model is as expected → leads to many conflicts and errors.
• Sidebar and route navigation is weird, so be explicit about that.
• Initial: All items in the sidebar fall under index. Data from one route doesn’t necessarily persist. e.g. If I connect an integration on the /integrations page, it doesn’t persist in the index page.
• Desire: All items in the sidebar have their own route and data persists across the app.

Logs

Ok, at this point, I am logging A LOT. There are too many errors and lots of back and forth so I’m only going to log highlights and learnings, at the top of this section to spare y’all the details!

Future