Query Logging and Feedback
Fetch query results
Query results store the query, chain-of-thought reasoning, answer, who asked it, and when.
Latest query result
JavaScript
lastQueryResult = Genai.Query.Result.fetch({ order: "descending(meta.created)", limit: 1 }).first()Filtered by user
JavaScript
Genai.Query.Result.fetch({ filter: Filter.eq("meta.createdBy", User.myUser().id) })Query count by user
JavaScript
c3Grid(Genai.Query.Result.eval({
"group": "meta.createdBy",
"projection": "meta.createdBy, count()",
"order": "descending(count())"
}))View user feedback on queries
Each piece of feedback is stored as a Genai.Query.Result.Feedback record and contains the following fields:
helpful(required boolean): whether the result was helpful (true= thumbs up,false= thumbs down)comment: optional free-text comment from the useradditionalFeedback: optional list of additional tags selected fromGenai.Query.Result.Feedback.Enum:- Positive tags:
Helpful,Accurate - Negative tags:
Not Helpful,Inaccurate,Outdated,Harmful - Response format tags:
Text Summary,Image,Table - Structured data tags:
Database Query,Visualization,Edited Structured Query Spec
- Positive tags:
Positive feedback
JavaScript
Genai.Query.Result.Feedback.fetch({ filter: "helpful == true"})Negative feedback
JavaScript
Genai.Query.Result.Feedback.fetch({ filter: "helpful == false"})Feedback with additional tags
JavaScript
Genai.Query.Result.Feedback.fetch({ filter: "additionalFeedback contains 'Inaccurate'"})Debug failures and poor answers
Engine Log
The engine log contains error messages and chain-of-thought reasoning for your query.
JavaScript
engineLog = queryResult.parsedEngineLogThe engine log shows the:
- Prompt passed to the LLM
- Error stack trace when a query fails
- Retrieved passages for unstructured data
- Post-processing steps to database query for structured data
Engine Log UX
Past query results and engine logs can also be viewed in the UI by selecting the History nav item. For more details, see Query Engine Log UX.
Improve quality of answers
Unstructured queries
- Add or remove documents.
- Add and embed document metadata in indexed content.
- Adjust LLM model parameters and prompts.
Structured queries
- Add or remove data sources.
- Edit table and column descriptions to accurately represent data.
- Add few-shot examples to the prompt for generating an eval spec:
- Add a sample question and spec for a query that received positive feedback.
- Add a sample question and corrected spec for a query that received negative feedback.