2.1 - Create a Knowledge Base

Creating Knowledge Bases for Chat Assistants

Objective:

To create a knowledge base for chat assistants to use as their primary source of information when providing answers to inquiries.

Key Steps:

1. Access the assistants menu and click on "Knowledge Base."

2. Create a new knowledge base by giving it a name.

3. Assign sources such as website data by adding new data sources.

4. Include text, FAQs (questions and answers), and files (Excel, Word, PowerPoint, PDFs) to the knowledge base.

5. Add relevant information and continue to update the knowledge base as needed.

Cautionary Notes:

• Ensure that the information added to the knowledge base is accurate and up-to-date.

• Regularly review and update the knowledge base to maintain its relevance.

• Double-check the sources and data added to the knowledge base for accuracy.

Provide context to your AI

Knowledge Bases - Detailed Document

Knowledge base are a collection of sources of information that your agent can access to retrieve relevant information during conversation. It can greatly improve the quality of the responses and the overall experience, especially in cases where there are a lot of information available (too long for putting in prompt), but having the information is essential for the agent to respond correctly. This feature quite useful for use cases like support, helpdesk, FAQ, etc.

 

Supported sources:

• Website content (via URLs)

• Documents (supported formats: .bmp, .csv, .doc, .docx, .eml, .epub, .heic, .html, .jpeg, .png, .md, .msg, .odt, .org, .p7s, .pdf, .png, .ppt, .pptx, .rst, .rtf, .tiff, .txt, .tsv, .xls, .xlsx, .xml)

• Custom text snippets

 

How it works

You can create knowledge bases, and link them to your assistants. When a knowledge base is linked to an assistant, the assistant will always try to retrieve information from the knowledge base before responding. There’s no need to change your prompt for it to trigger, as it will be done automatically, for every response generation.

During the creation of knowledge bases, it will chunk the source, embed them and store into a vector database.

During the conversation, when the assistant is about to respond, it will use the conversation so far (prompt is not included) to find the most relevant chunks from the knowledge base, and feed them to the LLM as context.

 

Best Practices

• Use markdown format for the source, with clear paragraphs: it will be easier to chunk and is easier to be retrieved.

  • For tabular data or image data, it’s sometimes hard for the system to recognize the correct data and put relevant information together in one chunk, so it might yield suboptimal results.

• Group relevant information together in one chunk: it will improve the retrieval accuracy.

• Avoid ambiguity and use specificity in references: when the chunk is retrieved, it’s best to contain specific references like name, dates, and less ambiguous terms like it or this. Because without the prior chunk, it’s hard to understand ambiguous terms.

 

---

 

FAQs

Do I need to prompt my assistant to use the knowledge base?

• No, you don't - it will be automatically fed as context to the AI during response. With that, referencing knowledge base in your prompt won't yield further usage of the knowledge base since the AI technically doesn't understand that it is being fed knowledge from a knowledge base, it just thinks there is context to use.

 

My assistant still isn't responding correctly, what should I do?

• Please check your knowledge base files to make sure all of the information is as correct as possible. Also, make sure to check the format of the source as suggest to make sure it can be properly read during embedding. Use the knowledge base testing environment to test and refine answers from your knowledge base.

 

Will this add to latency of responses?

• Theoretically - yes. Is it a noticeable amount? Depending on many factors, it could add 100 - 1000 ms of latency during a response.

 

Is there a way to check what is being used / retrieved during knowledge base calls?

• Yes! The knowledge base testing environment or in your conversations tab - click on the down arrow on a message and it will show the relevant knowledge pulled to generate that response.

 

Anything else that would help me?

• Turning your temperature (found in the autopilot settings of the assistant) to 0.00 will make the model more deterministic, in turn forcing the AI to follow instructions more accurately (I.e. the context of the knowledge base).