Reference only; verify against the target FastGPT release and deployment documentation.
What this category of data looks like
Data for this category comes from special equipment safety supervision systems, inspection and testing institution systems, user account books, and operator certificate databases. Daily documents include use registration certificates, periodic inspection reports, maintenance records, hidden danger rectification notices, accident reports, and operator certificates. The data update frequency is monthly, approximately 30 days. Document fields cover unique device codes, inspection cycle nodes, operator qualification validity periods, hidden danger rectification status, and more. Each document has an independent structure, with no unified cross-category general format. Each document type must be split and stored separately.
What constraints do these characteristics impose on the knowledge base retrieval and recall link
Data is dispersed across multiple systems. Multiple independent datasets must be associated to complete retrieval, which creates clear requirements for cross-collection recall capabilities. The monthly update frequency means retrieval results must match the latest inspection cycle and qualification validity data, to avoid using expired information. Core retrieval fields differ significantly across document types. For example, periodic inspection reports need to locate the next inspection node, while hidden danger rectification notices need to associate device numbers and rectification status. Differentiated retrieval and recall rules must be configured for each document type. Errors in supervision scenarios are directly tied to compliance and safety. Recall accuracy must cover core business fields to avoid missing key information.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Balances recall precision and coverage, and meets the compliance retrieval needs of special equipment supervision |
limit | Top 8 entries | Covers the number of associated documents required for a single document, and avoids exceeding the model's context processing limit due to overly long context |
searchMode | hybrid | Combines keyword and semantic retrieval, and adapts to scenarios with mixed structured and unstructured fields such as device numbers and inspection cycles |
usingReRank | Enabled | Improves the ranking priority of highly relevant results, and reduces missed detections of core documents such as hidden danger rectification notices and inspection reports |
datasetSearchUsingExtensionQuery | Enabled | Associates multi-source datasets, and covers different data sources such as safety supervision systems and inspection and testing institution systems |
collectionIds | Calibrated based on actual testing | Adapts to exclusive dataset collections of different supervision units, and avoids interference from irrelevant data in retrieval |
Three Common Mistakes
- Retrieval results only cover documents from a single system, and do not include report data from inspection and testing institutions. Cause:
datasetSearchUsingExtensionQueryis not enabled. Only datasets from a single data source are called, and cross-system core business documents are missed. - Historical call logs for a single document cannot be viewed. Cause: Log collection configuration is not enabled, or the log recording function is not bound to the module corresponding to document drafting.
- The error
The dollar ($) prefixed field '$schema'appears after deployment, or no corresponding configuration option is available when trying to enable the Graph RAG function. Cause: For the first case, the imported dataset configuration contains an unsupported$schemafield. Remove this field before re-uploading. For the second case, the current version does not have built-in Graph RAG support. Wait for subsequent version updates.
How to Confirm the Configuration Is Correct
- Manually enter the keyword for a specified device number, and check whether the retrieval results include the corresponding periodic inspection reports and maintenance records from the system.
- Navigate to the dataset management page, and confirm that all allowed data source collections have been associated, with no missing business account books.
- Navigate to the log query module, verify that call records for the specified document can be retrieved, and confirm that the log collection function operates normally.
- Trigger a document drafting process, and check whether recalled documents return results within 10 seconds and include the latest monthly updated data.
The values provided are common starting points and should be measured against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.