​

Knowledge Base Generator — Onboarding Screen Plan

​

Design Principles

1. Collect inputs, not reviews — onboarding gathers what the pipeline needs to run; the user reviews the output from the dashboard after generation completes
2. Don’t make them wait — generation runs in the background; get them to their dashboard fast
3. One thing per screen — each screen has a single clear purpose and a single primary action
4. As few questions as possible — only ask what the system cannot figure out on its own

​

Flow Summary

Screen 1: Business Identity
    ↓
Screen 2: Website URL
    ↓
Screen 3: Additional Knowledge Sources
    ↓
Screen 4: Launch → Dashboard

​

Screen 1 — Business Identity

​

Content

• Headline: “Let’s set up your AI”
• Subhead: “Tell us a bit about your business so we can build the smartest assistant possible.”
• Form fields (see table below)
• Primary button: “Continue”

​

Fields

​

Behavior

• No async work starts yet — this is pure data collection.
• Validation on “Continue”: all required fields must be filled.
• Selecting other for business type could optionally reveal a free-text field for the user to describe their industry.

​

Screen 2 — Website URL

​

Content

• Headline: “What’s your website?”
• Subhead: “We’ll scan your site to find your services, pricing, and brand voice — then use AI to build a complete knowledge base.”
• Single input field: URL (prefilled with https://)
• Helper text: “We’ll crawl your entire site. This works best with a live, public-facing website.”
• Primary button: “Continue”
• Secondary link: “I don’t have a website” — skips this screen (the pipeline will rely entirely on the additional sources from Screen 3 and manual entry from the dashboard)

​

Fields

​

Behavior

• No crawl starts yet. The URL is saved and passed to the pipeline on Screen 4.
• If the user skips, training.sourceUrl is set to null and the pipeline adjusts accordingly (heavier reliance on uploaded documents + manual dashboard entry).

​

Screen 3 — Additional Knowledge Sources

​

Content

• Headline: “Add extra knowledge”
• Subhead: “Got documents your website doesn’t cover? Upload them here so your AI knows everything. You can always add more later from the dashboard.”
• Three source options, displayed as a tabbed or segmented control:

​

Tab A — File Upload

• Drag-and-drop zone + file picker button
• Accepted formats: PDF, DOCX, TXT, MD, CSV
• Max file size: 25 MB per file

​

Tab B — Google Drive

• “Connect Google Drive” button (OAuth flow)
• After connecting: file/folder picker showing the user’s Drive
• Selected items appear as removable chips below the picker

​

Tab C — GitHub

• “Connect GitHub” button (OAuth flow)
• After connecting: repo/file browser
• Selected items appear as removable chips below the picker

​

Shared Constraint

• A shared counter is visible at all times: “X / 10 sources added”
• The limit of 10 is pooled across all three source types (any combination). This cap keeps generation time reasonable and reduces the chance of pipeline failure.
• When 10 sources are reached, all add/upload controls disable with the message: “You’ve hit the 10-source limit. Remove a source to add a different one.”
• All added sources are listed below the tabs in a unified list showing: source icon (file/Drive/GitHub), name, size, and a remove button.

​

Fields

​

Behavior

• This screen is entirely optional — the user can skip it.
• Primary button: “Continue” (always active, whether or not sources are added)
• Secondary link: “Skip — I’ll add these later”

​

Screen 4 — Launch

​

Content

• Headline: “Ready to build your AI”
• Quick summary card showing what was collected:
  • Business name + location + industry
  • Website URL (or “No website provided”)
  • Number of additional sources (or “None added”)
• Primary button: “Build My AI”
• Microcopy below button: “This usually takes 5–10 minutes. We’ll let you know when it’s ready.”

​

Fields

​

Behavior

• Pressing “Build My AI” triggers the full automated pipeline:
  1. Website crawl (Firecrawl)
  2. Entity extraction (business info, services, brand signals)
  3. Additional source ingestion and extraction
  4. Service enhancement (education, concern mapping, self-ID triggers, FAQs, differentiators — per service)
  5. Concern → service relationship mapping
  6. Conversation starter generation
  7. System prompt generation
  8. Assessment quiz generation
  9. Service guide compilation
• The user is immediately redirected to their dashboard — they do not wait.
• On the dashboard, a persistent status card shows real-time generation progress and updates to “Ready” with a CTA to review and test the AI when done.
• An in-app notification (toast/banner) fires when generation completes.
• An email is also sent: “Your AI assistant is ready.”

​

Screen Map at a Glance

​

What Happens After Onboarding

​

Data Model Addition

training: {
  sourceUrl: "https://example.com",   // or null if skipped
  crawledAt: "2026-03-10T...",
  pagesAnalyzed: 24,
  lastUpdated: "2026-03-10T...",
  version: 1,

  // NEW
  additionalSources: [
    {
      type: "file",          // "file" | "gdrive" | "github"
      name: "pricing-sheet-2026.pdf",
      reference: "/uploads/abc123.pdf",   // internal path, Drive file ID, or GitHub path
      sizeBytes: 245000,
      status: "processed",   // "pending" | "processing" | "processed" | "failed"
      addedAt: "2026-03-10T..."
    }
    // ... up to 10 items total
  ]
}