Configuring the job parameters and schedule for using Problem Solver

As a problem coordinator, you must configure the machine-learning job parameters and schedule their run to generate problem candidate clusters in the Console in BMC Helix ITSM.

Overview of Large Language Model parameters

The following Large Language Model (LLM) parameters are available in job configuration:

Parameter	Description
LLM Configuration	Defines the models used for clustering, cleaning, and enrichment of incident details. These models are used to process incidents for the capability.
General	Defines the lookback period, filters, and request-response time for the LLM.
Batch Config	Defines how the input incident details are split into manageable chunks for the LLM to process.
Extract	Defines how LLM pulls specific, structured details from incident details. It structures incident details into machine-readable form for further analysis.
Enrich	Defines how new details are added to the incidents. Enrichment may include summarization, classification, keyword generation, and more. It adds context, meaning, and intelligence to the original incident details.
Clean	Defines how the input incidents are cleaned and normalized before LLM can process them. Cleaning actions may include removing HTML, special characters, or noise. It may also include normalizing texts, removing duplicate and empty fields from incident details. It ensures high-quality, consistent input incidents for better results.
Cluster	Defines how incidents of similar details are grouped into semantically similar clusters. It ensures incident organization so that the LLM can reason over groups rather than isolated incidents.

Before you begin

You must have the Problem Config, Problem Master, or Problem User permission to configure and schedule jobs.

To set up the job parameters

You must configure the LLM-based job parameters for the algorithm to generate problem candidate clusters.

In BMC Helix ITSM, click .
On the Settings page, select Job Config > Problem Solver > Config.

In the following sections, enter values in the fields:

Section	Field	Description
LLM Configuration	Default model ID	Enter the unique identifier of the default model for the processing pipeline.
	Cluster Model	Enter the unique identifier for the clustering model used in the processing pipeline.
	Clean Model	Enter the unique identifier for the cleaning model used in the processing pipeline.
	Enrich Model	Enter the unique identifier for the enrichment model used in the processing pipeline.
General	Initial Lookback (Months)	Enter the lookback period (in months) up to which the algorithm considers the incidents for generating clusters. The default value is 3. The value applies to the first job run. For subsequent job runs, the algorithm does not consider the initial lookback period.
	Incident Filters	Enter status filters in JSON format to filter incidents. For example, {"Status": ["Resolved", "Closed"], "Company": []} is the default filter used to process resolved and closed incidents across all available companies. You can include company names in the filter to process the incidents from the selected companies.
	Request Timeout (ms)	Enter the maximum duration in milliseconds the algorithm waits for the server to respond before abandoning the request. The default value is 240000. We recommend that you do not edit this value.
Batch Config	Batch Threshold	Enter the number of incidents that trigger the algorithm's batch-processing action. The default value is 1000.
	Minimum Batch Size	Enter the minimum number of incidents that the algorithm considers in each batch for processing. The default value is 500. If a batch contains fewer incidents than the configured value, the algorithm does not process it.
	Max Batch Size	Enter the maximum number of incidents that the algorithm considers in each batch for processing. The default value is 10000.
Extract	Extract Max Records	Enter the maximum number of incident records that the algorithm extracts from BMC Helix ITSM. The default value is 100000.
	Use Batching	Select True. The default value is True. We recommend that you do not change the value.
	Extract Max Workers	Enter the maximum number of parallel workers the algorithm uses to extract the incident details. The default value is 30.
Enrich	Growth Factor	Enter the threshold value of the cluster growth after which the algorithm reperforms enrichment. For example, if the growth factor is 1.5, the algorithm re-enriches when the cluster size exceeds 50 percent.
	Enrich Max Incidents	Enter the maximum number of incidents to be selected from a pool of existing incidents for sampling as a part of the enrichment process. The default value is 15.
	Enrich Max Workers	Enter the maximum number of concurrent enrichment requests that the algorithm considers. The default value is 5. Increasing the value may increase the algorithm's processing time.
	Enrich Max Retries	Enter the maximum number of retry attempts made by the algorithm for failed enrichment requests. The default value is 3.
	LLM Max Attempts	Enter the maximum number of LLM call attempts that the algorithm can perform. The default value is 7.
	Enrich Batch Size	Enter the number of clusters that the algorithm can enrich per batch. The default value is 40. We recommend that you do not change the value of this field.
Clean	Clean Max Workers	Enter the maximum number of parallel workers the algorithm uses to clean the incident details. The default value is 30. We recommend that you do not change the value.
	Clean Max Retries	Enter the maximum number of retry attempts made by the algorithm for failed cleaning requests. The default value is 6. We recommend that you do not change the value.
	Cleaning Batch Size	Enter the number of incidents the algorithm processes per batch. The default value is 500. We recommend that you do not change the value.
	Cluster Max Attempts	Enter the maximum number of clustering attempts that the algorithm can perform. The default value is 3. We recommend that you do not change the value.
Cluster	Min Cluster Size	Enter the minimum number of incidents required to form a cluster. The default value is 10. We recommend that you do not change the value.
	Target Ave Size	Enter the average number of incidents per cluster. The default value is 15. We recommend that you do not change the value.
	Post Min Cluster Size	Enter the minimum number of incidents to include in each cluster. The default value is 20. We recommend that you do not change the value.
	Enable Quality Metrics	Select True. The default value is True. We recommend that you do not change the value.

Click Save.
(Optional) Click Reset to default to revert to the default field values.

To schedule a job in BMC HelixGPT Agent Studio

You must configure a schedule to generate problem candidate clusters in the console.

In HelixGPT Agent Studio, click .
Select HelixGPT > Connections > Information sources.
On the Information sources page, select Problem Solver.
In the Edit connection panel, in the Schedule section, enter the following details:
1. From the Month dates list, select the month's dates when the job must be run.
2. From the Days of the week list, select the days of the week which the job must be run.
3. In Time, enter the time of the day when the job must be run.
Click Save.
The job runs at the scheduled date and time, and generates problem candidate clusters in the console.

Where to go from here

Viewing the Problem Solver Console