Repository navigation
Assessment: Implement single response call #1327
Description
Is your feature request related to a problem?
The current implementation of POST /assessments only supports BATCH requests and returns a 501 error for single-row inputs. This limits flexibility and usability for users needing responsivity for single-row assessments.
Describe the solution you'd like
- Enable a single-row object in
input.datato run as RESPONSE while maintaining BATCH for a list of rows. - Ensure same row validation as in batch, including
input_schemaand URL-only attachments. - Implement a pre-filter gate followed by direct provider calls in one Celery task.
- Deliver results or failures exclusively via a required
callback_urlfor RESPONSE. - Reuse existing batch request builders, param mappers, and LLM-call clients with no migration required.
Original issue
Context
POST /assessments serves BATCH only; a single-row input returns 501.
Scope
- A single row object in
input.dataruns as RESPONSE; a list of rows stays BATCH. - Same row validation as batch: config
input_schema, URL-only attachments (image / pdf / video / audio,http(s)://orgs://). - Pre-filter gate, then assessment, run as direct provider calls (OpenAI, Anthropic, Gemini AI Studio, Vertex) in one Celery task.
- The result or the failure is delivered only by
callback_url, which is required for RESPONSE. - Reuses the batch request builders, the param mappers and the llm-call clients; no migration.
Out of scope
FAST mode, optional pre-filter / assessment, poll support for RESPONSE.
Metadata
Metadata
Assignees
Labels
Type
Projects
- Status
Activity
[-]Assessment: Add RESPONSE method (single row, result by callback)[/-][+]Assessment: Enable single-row requests[/+]feat(assessment): Add RESPONSE method with callback delivery (#1327)
[-]Assessment: Enable single-row requests[/-][+]Assessment: Implement single response call[/+]feat(assessment): Record RESPONSE runs and report pre-filter errors (#…