đź“š

How to Build an AI Knowledge Base for Your Podio App Using Claude Code

A step-by-step guide to creating LLM-ready documentation from your existing help resources and support tickets

Introduction

We recently built a comprehensive AI knowledge base for GlobiMail, our email integration service for Podio. The goal was simple: create documentation that an AI assistant could use to accurately answer customer questions without human intervention.

This guide walks you through the exact process we used, so you can do the same for your own Podio-based business. By the end, you'll have structured knowledge files ready to use with ProcFu's AI features or any other LLM-powered support system.

Why Two Separate Files?

We intentionally split our knowledge base into two distinct files:

1. llm_help.md - Product Knowledge & Instructions

This file contains:

  • Product overview and features
  • How the system works
  • Pricing and plans
  • Configuration guides
  • FAQ answers
  • Setup instructions

Why separate? This information is relatively stable. It changes when you update features or pricing, but not daily. It provides the foundational context an AI needs to understand your product.

2. llm_desk.md - Support Ticket Solutions

This file contains:

  • Real problem/solution pairs from resolved tickets
  • Troubleshooting steps that actually worked
  • Edge cases and their fixes
  • Common user mistakes and how to address them

Why separate? Ticket history is dynamic and grows over time. Keeping it separate means you can:

  • Update it regularly without touching your core documentation
  • Control the size and relevance of included tickets
  • Easily regenerate it from fresh ticket exports

Step 1: Gather Your Existing Documentation

Start by collecting all your current help resources:

Website Content

We used Claude Code to systematically visit every page on our website and extract relevant information.

Prompt used:

Visit all the pages on <http://www.yoursite.com> and update this document
based on what you find. Include an intro for the LLM explaining what
the product is and key points to remember when answering questions.

Help Manual / Knowledge Base

If you have documentation in Notion, Confluence, or another platform, export it:

  • Notion: Export as Markdown
  • Confluence: Export as HTML or PDF
  • Help desk software: Export articles as CSV or HTML

We exported our Notion help documentation as Markdown and had Claude merge it:

Prompt used:

Merge anything useful from docs/exports/notion_help.md into this document

Spreadsheet Exports

For structured data like FAQ lists, wiki articles, or canned responses, Excel exports work well:

Prompt used:

Merge anything useful from docs/exports/wiki.xlsx into this document

Step 2: Extract Your Support Ticket History

This is where the real value lies. Your resolved support tickets contain proven solutions to real customer problems.

Preparing Your Podio Help Desk App

To export complete ticket conversations (not just the item fields), you need to capture the comments. Here's how:

2.1 Create a Hidden Field for Comments

  1. Open your Help Desk app in Podio
  1. Add a new Multi-line Text field
  1. Name it something like "Communication Thread" or "All Comments"
  1. Set the field to Hidden Always (so it doesn't clutter your UI)

2.2 Create a PWA Flow to Capture Comments

Create a Podio Workflow Automation (PWA) flow that:

  1. Triggers manually or on item update
  1. Updates the hidden text field with “All Comments” - a token that’s available from the token picker

2.3 Run the Flow for All Existing Tickets

In PWA

  1. Select the flow you just created, and click the “Run” option
  1. Enter “0” for the item ID - so it will trigger on all items
  1. Wait - this process can take a while depending on how many items you have in the app.

This populates the hidden field for all your historical tickets.

2.4 Export to Excel

  1. Open your Help Desk app in Podio
  1. Make sure you’re on a view that shows all items or filter for resolved items
  1. Click the wrench icon > Export to Excel

You now have a spreadsheet with complete ticket conversations.

Step 3: Process Tickets into Problem/Solution Format

This is where AI assistance shines. Feed your ticket export to Claude and ask it to extract reusable solutions.

Prompt used:

The file /docs/exports/Help Desk Tickets.xlsx contains help desk tickets.
Each row represents a ticket. For each one, see if there's a definite reusable
solution, and merge it into llm_desk.md in a problem/solution format.

Claude will:

  • Parse the ticket data
  • Identify which tickets have clear, reusable solutions
  • Extract the problem description
  • Extract the solution that was provided
  • Organize them by category
  • Skip tickets that are too specific or don't have clear resolutions

Follow-up prompt to get more:

Can you add more? Add all relevant ones.

Step 4: Structure for LLM Consumption

Both files should start with an introduction that gives the AI context:

# [Product Name] Knowledge Base for LLMs

## Introduction for AI Assistants

This document serves as a comprehensive knowledge base for AI assistants
answering user questions about [Product].

Key points to remember:
- [Critical fact 1]
- [Critical fact 2]
- [Common misconception to avoid]

End with reminders for the AI:

## Important Reminders for AI Assistants

1. **[Key Process]** - Most issues are solved by doing X
2. **[Common Gotcha]** - Always check Y before suggesting Z
3. **[Policy Note]** - We don't offer phone support

Step 5: Using with ProcFu

Once your knowledge files are ready, you can use them with ProcFu's AI capabilities:

  1. Go to Account > Trained AI’s
  1. Copy & paste the content of both files to the training doc entry field. Leave a nice space between their respective content so you can find that mark again more easily later.

You now have a trained AI that you can call from PWA using the podio_ai_ask_trained function that can:

  • Answer product questions
  • Troubleshoot issues
  • Provide consistent, accurate responses based on your proven solutions

Useful Prompts for Building Your Knowledge Base

Here are prompts you can adapt for your own documentation:

For Website Content Extraction

Visit all pages on [your-website.com] and create a comprehensive
knowledge base document. Include:
- Product overview
- All features and how they work
- Pricing information
- Setup/configuration guides
- FAQ content

Format it for use by an LLM answering customer questions.

For Merging External Documentation

Merge the useful information from [exported-file] into this document.
Avoid duplicating information already present. Organize by category.

For Processing Support Tickets

This spreadsheet contains resolved support tickets. Column A is the
ticket title, Column B is the customer's question, Column C contains
the conversation thread with our response.

Extract problem/solution pairs that would be useful for handling
similar future tickets. Skip tickets that are:
- Too specific to one customer
- Missing clear resolutions
- About account-specific issues (refunds, cancellations)

Organize by category (Billing, Setup, Troubleshooting, etc.)

For Identifying Gaps

Review this knowledge base document. What common questions or issues
are NOT covered? What additional sections should we add?

For Quality Check

Review these problem/solution entries. Are any solutions:
- Outdated?
- Missing important steps?
- Contradicting information in the help document?

Flag any issues found.

Maintenance Tips

Regular Updates

  • Weekly: Export new resolved tickets and add valuable solutions
  • Monthly: Review and remove outdated information
  • On product changes: Update llm_help.md immediately

Version Control

Keep your knowledge files in Git so you can:

  • Track changes over time
  • Roll back if needed
  • See what was added/removed

Testing

Periodically test your AI with real customer questions:

  1. Take 10 recent support tickets
  1. Ask your AI assistant how it would respond
  1. Compare to how your team actually responded
  1. Identify gaps and add missing information

Summary

Building an AI knowledge base from your existing resources is straightforward:

  1. Gather documentation from your website, help center, and internal wikis
  1. Export support tickets with full conversation history from Podio
  1. Use AI to process and structure the information
  1. Separate concerns - keep product docs and ticket solutions in different files
  1. Connect to ProcFu for automated customer support

The investment pays off quickly. Every ticket your AI handles correctly is time saved for your team - and faster resolution for your customers.