To get the most out of Cassidy, it's important to follow best practices related to your data, prompting, limitations, and troubleshooting. This article will guide you through these key areas to help you achieve the best results.
Data:
Import relevant and authoritative documents: The quality of the AI's output depends on the quality of the input data. Ensure that you import documents that are relevant, authoritative, and up-to-date.
Maintain a clean knowledge base: Regularly review and update your knowledge base to ensure that the AI doesn't base its answers on outdated or irrelevant information. When creating assistants, limit the scope of each to only the parts of your knowledge base that are related (e.g., an internal assistant should have policy documents, handbooks, etc., but ideally not engineering documents). Cassidy also offers a document verification feature that helps clean up this data over time, see this article for more information.
Prompting:
Be specific in your queries: Provide context in your prompts and queries to help the AI better understand what information is most relevant. The more specific you are, the more accurate the results will be.
Ask questions about document content: Focus your questions on the content of your documents rather than metadata like dates and titles, as the semantic AI search does not take these into consideration.
Troubleshooting:
Rephrase and add context: If you receive a poor result, try rephrasing your question, providing more context, or guiding the AI in the right direction. This can help the AI better understand your query and provide more relevant information.
Use the appropriate assistant: Ensure that you're using the correct assistant for your needs. Common cases include:
For questions about Cassidy's features or how to use something, switch to the "Cassidy Help and Use Case Assistant."
For data analysis questions, use the dedicated "Data Analyst Assistant."
If you're asking questions about your knowledge base and not finding the information you need, consider whether you're using a Q&A assistant and if the information is within the assistant's knowledge base scope. If you know where the information is located, you can reference it directly using a hashtag.
For general questions where you want the AI to pull from its broader knowledge, use a non-Q&A assistant or a default assistant/model.
Clear chat: Try asking the same question in a new chat. As a chat becomes longer with many messages, the AI model can more easily be confused or forget details.
Limitations:
Metadata is not considered: The semantic AI search does not take into account metadata like dates and titles, so questions like "What was said in our last meeting?" or "Where is the file called XYZ located?" will not work.
No access to underlying document links: The AI does not have access to the underlying links for referenced documents, so if you ask it to output links to them, it may generate incorrect ones.
Text-based responses only: Currently, Cassidy only produces text-based responses and cannot generate images, audio, or other media.
By following these best practices and keeping the limitations in mind, you can optimize your experience with Cassidy and get the most accurate and relevant results for your needs. If you encounter any issues or have further questions, don't hesitate to reach out to our support team for assistance.