How to Build an AI "Brain"

The quality of your AI's responses is directly proportional to the quality of the information in your Knowledge Base. This article explains how to optimize your data sources for maximum accuracy.

Supported Data Sources

• Websites & Sitemaps: Ideal for public documentation. We recommend using a sitemap.xml URL to ensure the crawler finds every page automatically.
• File Uploads: Supports .txt, .csv, and .json. This is great for private internal data or "cheat sheets" you want the bot to reference.
• External Integrations: Connect Notion, GitBook, or Zendesk to pull in your existing help centers without manual copying and pasting.

How Ingestion Works

When you add a source, Apoyo performs three actions:

1. Scraping: We pull the raw text from your pages, stripping away unnecessary headers, footers, and scripts.
2. Chunking: We break long documents into smaller "concepts" so the AI can find specific needle-in-a-haystack answers quickly.
3. Embedding: We turn those concepts into mathematical vectors. This allows the AI to understand "context" rather than just matching keywords.

Troubleshooting Low Accuracy

If your bot is giving vague answers:

• Check the Match Threshold: In the Model settings, a lower threshold makes the bot "looser" (pulling more context), while a higher threshold makes it "stricter" (only answering if it is very sure).
• Format your Docs: Use clear headers (H1, H2, H3) in your documents. The AI loves structured data.