Control and Optimize LLMs for the Enterprise with the Mobile Coach Platform

The emergence of Large Language Models (LLMs) is rapidly changing the way organizations use artificial intelligence (AI). The ease in which LLMs help employees and customers get answers to questions, perform tasks, create content and receive overall guidance is driving wide adoption of this technology.

However, organizations need to manage this technology to ensure LLM usage is safe, compliant, and accurate. Just as organizations expect human staff to represent their company accurately, lawfully, and responsibly, an AI chatbot should likewise adhere to the same standards.?

The Mobile Coach Platform (MCP) is the ideal platform to help organizations harness and manage the power of an LLM. MCP is a robust rules engine that has many features to dictate what a chatbot should say and when. This rules engine can be used to create a “rules buffer” to manage LLM input and output.?

This rules buffer can be configured to do a number of impressive tasks:

1. Supercharge Personalization. Provide the LLM with detailed information about user attributes for a more personalized experience.
2. Dictate Content Sources. Provide the LLM the exact content set to draw from instead of relying on the public data.
3. Quality Control of LLM Responses. Filter an LLM response to ensure its quality before sending the message to the user.
4. Harness NLP Expertise. Utilize the LLM solely for its superior Natural Language Processing (NLP) capabilities.

--

Supercharging Personalization?

MCP allows users to define a wide range of user attributes such as language preference, job function, competency levels, goals, or whatever other attributes they want to define. When designing an LLM prompt, users can then include these user attributes they have defined in Mobile Coach custom fields.?

Dictating Content Sources?

One of the primary concerns organizations have with LLM usage is that LLMs are trained on publicly available content. Organizations want LLM functionality but need LLM to be trained on company sanctioned content while protecting the confidentiality and proprietary nature of such content. In addition, LLMs trained on public data create uncertainty around content credibility.?

The Mobile Coach Platform allows for organizations to feed an LLM with specific, proprietary content at the time of the user interaction. Mobile Coach users can do this in several ways:

• Use the Mobile Coach Platform Knowledge Base feature to upload its knowledge base articles, blog posts, job aids and any other company sourced content. Mobile Coach will then search the knowledge base based on the topic of the user’s query and pass that information directly to the LLM, thus ensuring that the LLM responds based on company content.

• Use the Mobile Coach Platform Local Table feature. This feature works similarly to the knowledge base feature except performs better when the nature of the content is more of an Frequently Asked Question (FAQ) use case.?
• Include a specific article or company document in the prompt for a specific interaction. If the topic of the interaction is specific enough, users can include a source document in the prompt itself.?

Quality Control of LLM Responses

In some cases, organizations do not want to automatically pass an LLM response to a user without some kind of filtering or checking for quality. It is possible to set up an interception rule to parse the response with some alternate logic. For example:

Harnessing NLP Expertise?

Mobile Coach users can leverage LLMs and the Mobile Coach Platform to intelligently deal with natural language processing, within the context of the user interaction. Users configure the Mobile Coach Platform interface to pass the user’s text along with detailed instructions.

Examples:

This is how this looks in the platform: