Chat Limitations

We're continuously working to improve the quality of the responses. Here are the current limitations.

1. Data Availability

Lumi cannot answer questions requiring data it isn't connected to or aware of in the knowledge base. Questions beyond this scope may not receive accurate responses. You can check what data is available to query in the table preview or in the knowledge base.

Table Preview Feature

2. Spelling of Attributes

Misspellings in attributes (e.g., product names, brands, or categories) may lead to empty responses. If unsure, you can:

• Perform a general search: "how many items fall under the category that contains any variation of bikes in the description?"

• Add common abbreviations or variations to the knowledge base.

❌ Misspelling Example

Scenario: User asking about metrics about 'carbon road bike' sales

Error: There are no items that have the name 'carbon road bike' this will return nothing because Lumi is directly filtering for items with the explicit name 'carbon road bike'.

✅ Tips to Resolve Error

Option 1: Use Exact Spelling

Use the exact attribute name.

Example: What was the revenue of Carbon Elite Road Bike last year?

Option 2: General Search

Use language that generalizes the search criteria.

Example: What was the revenue items containing any variation of 'carbon' last year?

Option 3: Use Two Prompts

1. Perform a separate query to obtain all potential variations.

Example: Can you show me all the items that contain any variation of 'carbon' in the name?

2. Re-contextualize the prompt with the exact name

Example: What was the revenue of Carbon Elite Road Bike last year?

3. Follow Up Questions

Lumi includes follow-up questions in responses to encourage further user engagement. Please note that some of these follow-up questions might not always be directly relevant.

4. Data Returned From Query

To ensure that raw data is handled correctly, Lumi does not have access the the previous query data results. This means that prompts need to be structured with this in mind and should not reference data returned. Rather a rephrased prompt can be used to target whatever insight is desired.

5. Query Row Limit

To ensure unnecessary computation resources does not occur, a return limit is enabled to 25 records. While this can be modified it is important to be aware for regular use and common use cases.