GoHavi GoHavi / Docs

The Brain (AI Knowledge Base)

Train your AI Agent to answer customer queries accurately, 24/7.

In this guide

1. How it Works (RAG)

GoHavi uses a technology called Retrieval Augmented Generation (RAG). Unlike generic ChatGPT, our AI does not just "make things up." It follows a strict 2-step process:

  1. Retrieval: When a customer asks a question (e.g., "What is the price of a root canal?"), the system first searches your Knowledge Base for relevant text snippets.
  2. Generation: It then feeds those specific snippets to the AI (Gemini) with a strict instruction: "Answer the user using ONLY this information."
![Diagram: User Question -> Search Knowledge Base -> Context -> AI Response]

Figure 1: The Context Injection Flow

2. Training Your Brain

To make the AI useful, you must feed it Facts. Go to AI Studio > Knowledge Base to add content.

Best Practices for Facts

✅ Do This

  • "Our office hours are Mon-Sat, 9 AM - 6 PM."
  • "The price for X Service starts at $50."
  • Use short, standalone paragraphs.

❌ Avoid This

  • Uploading massive PDFs (currently not supported).
  • Vague statements like "Call us for price."
  • Conflicting information in different notes.

3. Important Limitations

While powerful, the AI is not a human employee. Understanding its limits is key to success.

1. Context Window

The AI can only "read" a certain amount of information at once. If your Knowledge Base is huge, it might miss a detail if the search algorithm doesn't find the exact right snippet first.

2. "I Don't Know" Policy

We strictly instruct the AI NOT to hallucinate. If the answer is not in your Knowledge Base, it will say:

"I'm not sure about that. Let me connect you with a human agent."

This is a safety feature, not a bug.

3. Structured Data

The Brain is great at text questions but struggles with complex math or dynamic lookups (e.g., "Is the 5 PM slot open?") unless integrated specifically with the Appointment Module.

4. Emotional Nuance

The AI may miss sarcasm or deep emotional frustration. We recommend enabling Sentiment Analysis (in Analytics) to flag angry customers for human intervention.