Intent detection (Advanced Access)

Advance Access (AA) is an opportunity to evaluate and work with Relativity features prior to the General Availability (GA) release. Relativity customers typically participate in AA programs on a feature-by-feature basis. The functionality described in this topic may not be available in all Relativity environments. This topic may not represent the functionality, appearance, or behavior of the GA release version of this feature.

Intent detection is one of the mechanisms that aiR Assist uses to understand the purpose behind each prompt and respond in the most appropriate way. Instead of simply matching keywords, aiR Assist analyzes the intent of a query to determine whether it should:

Retrieve and summarize information from indexed documents.
Ask for clarification when the question is unclear or refers to multiple topics.
Indicate when a request cannot be fulfilled due to missing data or unsupported actions.

This capability is designed to help aiR Assist provide relevant and context-aware answers, which may contribute to a more natural and efficient chat experience.

How intent detection works

When a question is submitted, aiR Assist automatically evaluates the likely intent and determines an appropriate response approach. Depending on the context and available data, one of the following response strategies is used:

Response strategy	Description	Recommended actions
Retrieval-based answering (using RAG)	When a question can be answered using the indexed documents, aiR Assist retrieves the most relevant information and generates a concise, well-supported response.	No action needed.
Refusal due to missing data	If the system cannot locate relevant information within the indexed dataset, aiR Assist will clearly indicate that no matching content is available.	Verify that the appropriate index is selected. Confirm that the expected documents are included in the saved search used to build the index. Rephrase the question to be more specific (for example, include names, dates, or topics). If information truly does not exist in the indexed content, no further action is required.
Refusal due to unsupported request	When a question asks for something beyond aiR Assist’s scope - such as legal opinions, drafting documents, or making recommendations - the system identifies the intent and responds appropriately.	Rephrase the request as a document-based question (such as, “What documents discuss termination terms?” instead of “Should we terminate the contract?”). Focus on facts, references, or evidence rather than conclusions or advice.
Disambiguation (clarifying ambiguous queries)	If a question references multiple possible entities or events, aiR Assist will prompt for clarification before continuing.	Provide the requested clarification.
Fallback for Unclear Intent	If the system cannot confidently determine what the question refers to, it will ask a clarifying question. If clarification is not provided, aiR Assist will default to generating a retrieval-based response using the available dataset.	Add relevant detail such as people, timeframe, or subject matter. Break broad questions into smaller, more focused inquiries. Confirm whether the goal is to find facts, summarize documents, or surface key themes.