‎Discovery of Generative AI FAQ Bot

Discovery of Generative AI FAQ Bot

Updated

19 days ago

This article will serve as a guide to discovering the requirements for building a Generative AI-powered FAQ Bot.

Identify your requirements for the FAQ Bot. Depending on your needs, two main use cases can be as follows:

Fully FAQ-Based Bot: If your requirements are to have an FAQ Bot powered by Generative AI that answers questions from a pre-defined knowledge base.
Journey-Based Conversational Bot: If you need a mix of transactional journeys and answering FAQs, the bot will be designed to not only respond to FAQs but also manage and promote user engagement along specific transactional paths. This can be implemented in two ways:
- Selecting a bot menu that engages the Generative AI-based bot.
- Running on the fallback messages.

Determine the number of languages in which the bot is expected to respond. Each language should be treated as a separate workflow, and the build time for each language will be planned accordingly.
First Preference for language: Both KB Article and our primary language for the bot is English.
Second Preference for language: KB Article is in English and primary language for the both is a different language.
Third Preference for language: Both KB Article and our primary language for the bot is same but a different language than English.

Sync from Knowledge Base
- If you have dynamic content stored in your CMS/KB, it needs to be imported into Sprinklr KB and synced at regular intervals.
- Customer's Website Integration: To pull data directly from the customer's website:
  - Integrate your Website Content Management System with Sprinklr KB.
  - Set up a Sync with KB.
Upload Documents: Static data in Excel, Doc, PPT, PDF, or a Zip file format can be uploaded directly from the local system.
Custom Question Answer Pairs: Define custom information in a Q&A format.

Coherent and Logical Flow: Ensure that the training data, whether in text or tabular format, has a coherent and logically flowing language. It should read like a properly written piece of text, making sense, flowing naturally, and easy to understand.
Minimal Columns in Tabular Data: Tabular data with numerous columns may result in lower accuracy due to the structural similarity between entries, which reduces semantic search results. It's advisable to include only necessary data, using as few columns as possible. If using multiple columns in an Excel file, ensure they provide additional information beyond the existing text. For example, adding a column for topics or themes can provide additional context while training.

What should be included?

Compilation of user queries along with preferred responses (optional).
Determine the number of user queries to be included in the test set.
Ensure the test set encompasses the breadth of content found in the training data, covering all possible query combinations.

Get started by building your own Generative AI powered FAQ bot!