One AI Assistant Generative AI Usage

General Information

One AI Assistant uses OpenAI large language models (LLMs) through One Model’s enterprise-grade OpenAI account. When you submit a prompt, only the prompt text and associated embeddings are sent to OpenAI, where they are stored for 30 days for logging and monitoring purposes. These inputs are not used to train OpenAI's models. Any fine-tuning is performed exclusively using synthetic examples generated by One Model. No customer data is used for training.

Usage of Generative AI in One AI Assistant

Generative AI is used for the following tasks in One AI Assistant:

Context Identification

An LLM task reviews the submitted prompt and determines whether the intended result is a visualization or a Storyboard. As we expand One AI Assistant's capabilities, this task will handle more "orchestration" of prompt intent.

Embeddings

One AI Assistant uses OpenAI’s embedding service to generate embeddings for user defined metrics, dimensions, and dimension nodes (e.g., Headcount, Gender, Location). Embeddings are numeric representations of words or phrases that help computers understand meaning and relationships between concepts. Once generated, these embeddings are stored in a vector database, hosted securely using pgvector within the One Model AWS environment. They are used in One AI Assistant to match terminology of a similar concept, such as matching a request for “Exits” to a “Terminations” metric.

Identification of Metrics, Dimensions, and Time Selections in Prompts

When a user submits a prompt, such as “Show me headcount by cost center for the last 12 months,” the prompt is sent to OpenAI. A fine-tuned LLM is leveraged to identify essential elements, such as metrics, dimensions, dimension nodes, and time selections within the prompt. For instance:

• Headcount is a metric
• Cost Center is a dimension
• Brisbane is a dimension node
• Last 12 months is a time selection

These recognized elements are then sent back to One Model, where One AI Assistant searches your vector database to find the closest matches in your data. Dimension nodes (e.g., “Engineering” within Cost Center) are also available in this process, allowing for specific filtering requests, such as “Show me headcount for the Engineering department only.”

Only data configured and permissioned for the assistant’s use during setup can be accessed, matched, and used in analysis. If a metric or dimension name was not embedded, it won’t be included in the assistant’s analysis, nor will it be shared with OpenAI.

One AI Assistant Insights

Ranking: After a chart or table is generated, if a user clicks “Show Insights,” One AI Assistant may display up to five insight statements. While the insights themselves are not generated by an LLM, the statements are ranked by an LLM based on their relevance and interest. Only the text of each insight is sent to OpenAI for ranking—the underlying data powering the chart or table is not shared at any point during this process.

Recommendations: The insight statements are sent to a separate LLM that generates a business recommendation for each one. Only the insight statement is sent to OpenAI.