To optimize use of the elvex platform, it's crucial to understand effective data management techniques. This includes knowing when to use the rules and context fields in the assistant instructions and when to use Datasources.
1. When to use Rules:
Rules in elvex primarily guide how an assistant should respond to inputs. They help in tasks such as content generation and formatting, but are not designed to hold factual information that the assistant needs to refer to.
For instance, if you're creating an assistant that generates customer support responses, you might set up a rule to ensure the content is always formal and polite, by disallowing any informal language or slang. This ensures a consistent professional tone in the assistant's responses without requiring the assistant to reference a fact or piece of data.
2. The Purpose of the Context field:
Contextual data provides specific knowledge the assistant can leverage to generate relevant responses. It assists in a generalized environment rather than to answer specific inquiries. It's best used for company or industry-specific knowledge like acronyms, internal lingo, or product names to ensure the assistant has the correct understanding of the terms it is likely to encounter in user requests and / or Datasources it will reference.
For instance, if you're creating an assistant for internal use within a pharmaceutical company, you might populate the Context with acronyms and jargon frequently used in the industry. This also might include the names of specific products or research projects within the company.
This way, when a user interacts with the assistant, it is equipped to understand and respond to queries in a way that makes sense within the specific pharmaceutical context. Providing the assistant with this context helps it generate more informed, accurate, and relevant content for users within that company or industry.
3. Role of Datasources:
Datasources furnish intelligence for assistants to complete tasks, making them a suitable repository for specific facts an assistant might need. Datasources are shareable across assistants - any assistant connected to the datasource can leverage the information therein.
The other important note is that assistant instructions (Rules and Context) are send to the LLM with each new user request on that assistant. This means we're taking up valuable space in the LLM's context window to provide it with facts that may not be at all relevant to the uer's request. By relying on Datasources to provide the right knowledge to the LLMs, we're being more resource efficient, improving performance and potentially cost as well.
For instance, suppose you have constructed an assistant designed to assist employees of a large organization in finding information about company policies. In such a scenario, the company's policy handbook could be turned into a datasource.
Then, when an employee asks the assistant a question like "What is the company's policy on parental leave?", the assistant could directly reference the datasource (the policy handbook), find the necessary information, and provide a precise, helpful response.
This way, datasources enable your assistant to provide enhanced intelligence based on the specific information it holds, making it invaluable for tasks requiring data-based responses.
The decision between using a rule, context, or datasource depends on your specific needs. However, updating the datasource typically ensures greater accessibility and wider application, enhancing the user experience. Careful and strategic use of elvex's features improves overall system efficiency on the elvex platform. If you need help making the right decision for your app, email us at [email protected] and we'll be glad to help!