adm_ai_job_automations_api

This document contains the API documentation for the adm_ai_job_automations_api package.

Functions and Procedures

sync_all_rag_collections

Synchronizes RAG collection files for all active RAG collections. Iterates over each collection and calls adm_ai_rag_workflow_api.sync_rag_collection_files.

Signature:

procedure sync_all_rag_collections (
  po_error_occurred out nocopy boolean
);

Parameters:

Name	Direction	Type	Description
po_error_occurred	out	nocopy boolean	Output parameter indicating if an error occurred during synchronization

---

process_rag_queues

Drains the RAG pipeline: extracts text, chunks it, embeds the chunks, and removes the vectors of files that have left a collection.

This is the half of the pipeline nothing used to run on a schedule - hourly_job fills the queue, and until this existed only the Sync Collection button emptied it, so a document became searchable when an administrator pressed a button.

Paced by AI_JOB_BATCH_SIZE (the batch, and the commit unit) and AI_JOB_MAX_SECONDS (the whole run). Work it does not get to is still queued for the next run, so stopping on the budget is a normal outcome rather than a failure. Failures are logged and reported through po_error_occurred; they do not abandon the other stages.

Signature:

procedure process_rag_queues (
  po_error_occurred out nocopy boolean
);

Parameters:

Name	Direction	Type	Description
po_error_occurred	out	nocopy boolean	Output parameter indicating if a stage failed during the run

---

hourly_job

Main AI hourly job procedure that orchestrates all AI-related automated maintenance tasks. This procedure is typically scheduled to run once per hour.

Signature:

procedure hourly_job;

---

queue_job

Entry point of ADM_AI_RAG_QUEUE_JOB, which runs every five minutes and drains the RAG pipeline through process_rag_queues.

Separate from hourly_job, and far more frequent, because the two halves cost very different things: discovery re-evaluates every document against every collection’s criteria, while draining the queue is bounded work on rows that are already there.

Signature:

procedure queue_job;

---