Enable Ask Relevance on Text or PDF

To be able to search through your data via an AI engine, your data needs to be in a format that AI understand it. Relevance AI has provided you with an easy workflow to make your data searchable. This workflow is the first step for building a chatbot based on your knowledge-based.

How to prepare data for a chatbot widget

Have your knowledge base in a conventional format such as CSV or PDF. Once you have uploaded your data, select your dataset, depending on your data type click on Enable Ask Relevance on Text or Enable Ask Relevance on PDF under Workflows and follow the instruction. If it is not listed on the Overview page (i.e. under "Suggested workflows for this dataset"), one easy way to access a workflow is to search for it under Browse Workflows as shown in the image below.

Setup steps are:

1. specify the field
   1. you wish to use for processing (if the source data is a CSV)
   2. in your dataset containing the URL to your source file (if the source data is PDF)
2. select your preferred chunking/splitting size

Chunking sizes specify the size of blocks of text to be used when looking for answers and retrieving the best matching answer to a question. So, for example, if your dataset contains short sentences as sources of truth chunking/splitting by sentences works fine. But if your need a paragraph to provide a meaningful answer use the recommended way.

3. Optional: review the optional settings

• Change to No, if you do not wish to receive and email notification upon workflow completion
• Set up an advanced filter so the workflow only processes a subset of the data
• Apply to the whole dataset means the platform automatically checks if there are unprocessed entries and only include them
• Include a link to your company logo for it to appear on the chatbot page
• Enter a title for your chatbot
• Sliding window is relevant to AI vector search and has to do with the number of tokens included in contextual similarity assessment.
• The maximum number of tokens in text splitting is relevant to chunking a page into its composing components.
• Model: If you do not change this option, by default Relevance AI's vectorizing model will be used for processing your data. The other two options are Open AI and Cohere AI. See How to use external API keys if you wish to use your own Open AI account.

4. Click on "Run workflow"

When the workflow is finalized, under history you can access links for testing the Ask Relevance feature in a chatbot widget or under the designed Relevance AI dashboard. Read more about it here.