Access Attributes Extraction
Purpose
Automatically extracts “access attributes” from document metadata/tags and attaches them to KBs so RAG / AiseraGPT can enforce content‑level access controls at query time.
When should I run this?
Run (or re‑run) after new KB content is ingested or access‑control tags/attributes are changed so that RAG responses respect the updated access model.
Unresolved Conversations Categorization
Purpose
Automatically categorizes unresolved conversations into high‑level reasons (e.g., Not Understood, KB Gap, Flow Gap) so teams can quickly see why conversations failed.
When should I run this?
Run on a recurring basis (e.g., daily/weekly) after you have enough new traffic, especially before analysis/review sessions, to ensure the latest unresolved conversations are categorized.
Unresolved Conversations Clustering
Purpose
Groups unresolved/“Not Understood” utterances into semantic clusters (topics) so admins can fix many similar issues (create or map intents) in bulk.
When should I run this?
Run after a batch of new conversations has accumulated (e.g., weekly or ahead of tuning cycles) to identify top unresolved topics and then mass‑attach them to intents from AI Workbench.
.Learning. Jobs (Conversation / Intent Learning – v1 vs v2)
Purpose
Continuously learn from conversations and training data to improve intent detection, ranking, and mapping of user phrases to intents, especially in the 2.0 conversation model.
When should I run this?
Run after a meaningful set of configuration or data changes: new/updated intents, large batches of labeled utterances, or significant changes in conversation patterns (e.g., post‑go‑live, after new use cases).
Relation Extraction (Deprecated)
Purpose (historical)
Extracted semantic relations between entities in KBs and stored them in the Knowledge Graph to support advanced reasoning.
When should I run this?
You typically should not run this anymore; it is deprecated and its use is replaced by newer RAG/neural search‑based capabilities.
App Content Classifier (ACC) (Deprecated with 2.0 routing)
Purpose
For universal bots, trained a classifier to decide which child bot/app should handle a given request based on app‑specific content and domain definition files.
When should I run this?
Only run in legacy/universal‑bot setups that still rely on ACC; for 2.0‑based and LLM‑driven routing, you do not need to run this, as routing is handled by newer domain detection and RAG‑based logic.
CPDC Training (Deprecated)
Purpose (historical)
Built domain classifiers (e.g., IT vs HR vs Facilities) from domain definition files and example content, primarily for multi‑domain/universal scenarios.
When should I run this?
In most 2.0 / AiseraGPT deployments you should not run this; only consider it for older, CPDC‑dependent configurations that have not yet migrated to LLM‑based domain detection.
Contextual Disambiguation (CDS) Training
Purpose
Supported models that use context (history, entities, configuration) to resolve ambiguous or overlapping intents/answers.
When should I run this?
For modern RAG/2.0 setups, you typically do not need to run this manually; contextual disambiguation is handled implicitly as part of RAG indexing and conversation model updates.
Entity Validation
Purpose
Validates configuration quality for flows, intents, and ontology, catching structural and schema issues (e.g., invalid flows, broken KB references, empty entity classes) before they cause runtime errors.
When should I run this?
Run after configuration changes in flows/intents/ontology (especially before go‑live or major releases) to proactively detect and fix configuration errors.
IAS Training
Purpose
Trains and refreshes internal AI services at tenant level (shared classifiers, signals, and models used across multiple features).
When should I run this?
Normally you do not need to schedule this manually; it is a platform‑managed system job, and if support advises a run, it would be in specific troubleshooting/tuning scenarios.
Core learning / model jobs
KGNER / NER Training
Purpose
Trains the KGNER/NER model on the tenant’s ontology and entities so the system can correctly recognize and tag entities in user utterances and tickets.
When should I run this?
Run after significant ontology changes (new/edited entities/categories) or after Ticket Learning / AI Learning setup, or when instructed as part of an offline retraining cycle.
Incident Model Job (Incident / Major Incident Clustering)
Purpose
Trains and scores models on historical incidents/alerts to detect patterns, cluster similar incidents, and support Major Incident identification in AI Observability.
When should I run this?
Run after you have a good volume of recent incidents (and after ticket ingestion config is stable) or when incident patterns change significantly (new apps, infra changes), and before enabling/refreshing Major Incident views.
Knowledge & fulfillment jobs
KB Learning / Knowledge Learning
Purpose
Learns from KB articles to recommend fulfillments (KBs) for existing intents and improve content‑based matching for search and answers.
When should I run this?
Run after KB ingestion is completed or updated, and after intents are reasonably stable (post Ticket/Conversation Learning), especially when you want the system to auto‑suggest the best KBs for each intent.
Indexing / retrieval jobs
Neural Search RAG Indexing / RAG Indexer
Purpose
Builds or refreshes the vector / RAG index used by AiseraGPT and Neural Search so that queries retrieve the latest content with semantic (embedding‑based) search.
When should I run this?
Run after KB or document ingestion changes (new/updated/deleted content), after enabling/disabling data sources, or after major RAG/Neural Search configuration changes; typically scheduled daily or a few times per day for active tenants.
Neural Search Flow Indexer / Workflow Indexing
Purpose
Indexes workflows and hyperflows so they can be retrieved and ranked by Neural Search (e.g., for flow search, agent assist, and RAG reasoning over flows).
When should I run this?
Run after creating or significantly updating flows/hyperflows, especially before testing flow search or any feature that surfaces flows via search; schedule periodically if flows change often.
Intent / conversation model jobs
ICM Training (Beta) / ICM v2 Training
Purpose
Trains or refreshes the intent classification model (ICM) so user queries are correctly mapped to intents based on the current bot configuration, ontology, and training phrases. “ICM Training (Beta)” is used for Conv 2.0 apps; “ICM v2 Training” is for Conv 1.0 stack.
When should I run this?
Run after adding/editing intents, updating training phrases, or making significant ontology/use‑case changes; also run as part of AI Lens “Retrain” workflows when you tune the intent model.
Utterance Validation
Purpose
Quickly checks if a candidate utterance is consistent with a configured intent and does not collide with or get misclassified into other intents before you approve/commit it.
When should I run this?
Use whenever you add or bulk‑import new training phrases to validate them before committing, especially on high‑traffic or sensitive bots where intent overlaps are risky.
(No single canonical doc, but this aligns with how utterance‑level validation is used around ICM training and AI Lens.)
Service Catalog indexing
Neural Search Service Catalog Indexer
Purpose
Builds/updates the Neural Search index for Service Catalog items (e.g., ServiceNow or other catalog systems) so catalog requests can be retrieved via neural (embedding‑based) Service Catalog Search, instead of keyword only.
When should I run this?
Run after Service Catalog learning / data source jobs complete, after enabling Service Catalog in fulfillments, or whenever catalog items are significantly updated; consider scheduling regularly to avoid missing indices and 500 errors in /rag/search.