What data for this service category looks like
Data for operations outsourced on-site service typically comes from the ITSM system, monitoring platform, asset inventory, and duty records. Update frequency is mostly daily or monthly. Daily generated documents include on-site daily reports, service desk work orders, inspection reports, and SLA monthly reports. Document fields vary across types: work orders contain number, processing status, and customer feedback; on-site daily reports contain on-site staff information and daily inspection items; inspection reports contain device numbers and exception records; SLA monthly reports contain compliance statistics items.
What constraints do these characteristics impose on reranking and hybrid retrieval?
Multi-source heterogeneous data sources require hybrid retrieval to cover field formats across different systems, to avoid matching keywords only for a single document type. Daily updated duty records and work orders require real-time retrieval capabilities, to ensure on-site engineers can access the latest on-site information. Business weights differ across documents: for example, urgent work orders should have higher priority than regular inspection reports. Therefore, the reranking link must adapt to the business attributes of different documents, to avoid irrelevant content interfering with on-site processing. Additionally, on-site service delivery relies on contractual SLA requirements, so retrieval results must prioritize fields related to SLA, to ensure the accuracy of statistical summaries.
How to set the configurations
The following configurations are formulated based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
usingReRank | Enabled | Adapts to the sorting needs of multi-source heterogeneous documents, and improves retrieval relevance for operations work orders and inspection reports |
similarity | 0.72–0.85 | Filters low-relevance on-site daily reports and inspection records, to avoid irrelevant information interfering with work order processing |
limit | Top 10 entries | Matches the upper limit of single retrieval results for daily operations processing, to control context length |
searchMode | hybrid | Supports hybrid retrieval across ITSM and monitoring platforms, covering all business data for on-site services |
Three common misconfigurations
- Phenomenon: The deployed reranker model is not displayed in the FastGPT interface. Cause: The model access credentials were not correctly configured in the configuration file, or the container network was not connected to the reranker service port.
- Phenomenon: Retrieval results are not sorted by work order priority. Cause: The
usingReRankconfiguration was not enabled, or thesimilaritythreshold was set too low, causing irrelevant results to be included. - Phenomenon: Reranked results are not displayed in the conversation details. Cause: Logging for reranked results was not enabled, or model call timed out without capturing returned content.
How to confirm the configuration is complete
- Enter the knowledge base configuration page in FastGPT, confirm that
usingReRankis enabled, and that the access address and port configuration for the reranker model are correct. - Initiate a simulated operations work order query, and check whether the sorting of retrieval results follows the business priority logic of on-site work orders.
- Check the conversation details page, confirm that reranked results are displayed normally, with no empty fields or error prompts.
- Initiate a cross-source retrieval test, retrieve relevant content from both ITSM work orders and duty records at the same time, and confirm that the hybrid retrieval function is working properly.
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 2026-10-19.