What the data for this category looks like
The knowledge base data for new hire onboarding mainly comes from official internal documents such as company internal onboarding guides, department job descriptions, system operation manuals, and attendance and social security process explanations. The data update rhythm is flexibly adjusted according to the company's onboarding plan, and full updates are usually completed before new batches of employees start work. Most documents are structured and chunked content, categorized by department, job position, and handling process, and include fields such as process steps, permission instructions, and contact information of liaisons. Common units involved include workdays, handling time limits, and liaison departments.
What constraints these characteristics impose on the knowledge base retrieval and recall link
The knowledge base data for new hire onboarding is updated temporarily according to the company's onboarding plan. This requires the retrieval link to support on-demand document synchronization to avoid retrieving expired process information. Documents are categorized by department, job position, and process, and new employees in different positions have retrieval needs focused on corresponding categorized content. Precise matching of category tags is needed to filter irrelevant information. Most documents contain complete process steps and liaison information, so retrieval results must cover the full process without redundant fragments, avoiding returning disconnected operation instructions. At the same time, data updates have no fixed cycle, so a flexible synchronization mechanism must be adapted to ensure that new hires always obtain the latest version of information.
How to Set the Configurations
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches precise process descriptions for new hire onboarding, avoids retrieving guidance content for unrelated positions, and adapts to the vector matching logic of FastGPT v4.17.0 |
limit | Top 6 entries | Covers the necessary steps of a complete onboarding process, while controlling the context length to avoid exceeding model limits |
searchMode | Hybrid vector search + keyword search | Balances process keyword matching and semantic similarity, accurately retrieves documents containing specific steps, and adapts to the retrieval needs of multi-step processes |
usingReRank | Enabled | Re-ranks retrieval results to prioritize content most relevant to the current onboarding scenario, such as administrative processes over technical training |
datasetSearchUsingExtensionQuery | Enabled | Optimizes colloquial questions from new employees, for example, automatically expanding "how to collect employee ID cards" to "employee ID card collection process, location, required materials", improving retrieval accuracy |
collectionIds | Corresponding dedicated dataset ID for onboarding guidance | Only retrieves knowledge base collections related to new hire onboarding, avoiding mixing in other business data |
Three Common Misconfigurations
- Phenomenon: The retrieved onboarding process content in the knowledge base does not match the latest updated static page. Cause: The on-demand knowledge base synchronization mechanism is not configured, and fixed-cycle synchronization is still used, resulting in the knowledge base not pulling the latest content after the static page is updated.
- Phenomenon: When asking questions consecutively in the same session, answers from the second question onward are off-topic. Cause: Context association configuration is not enabled, or the
limitvalue is set too high, causing redundant content to occupy the context and preventing retention of scenario information from previous questions. - Phenomenon: Retrieval results only return a small amount of fragmented content and do not cover the complete onboarding process. Cause:
searchModeonly uses pure vector search, without combining keyword matching, making it impossible to retrieve document fragments containing specific process keywords.
How to Verify Proper Configuration
- Upload an updated onboarding process static page, manually trigger knowledge base synchronization, and verify that the retrieval results include the latest content.
- Simulate consecutive questions: first ask "how to complete onboarding procedures", then ask "what materials need to be prepared", and verify that the answer links back to the process scenario of the previous question.
- Adjust the
similarityvalue, verify the accuracy of retrieval results under different thresholds, and confirm that they meet the matching requirements of the current scenario. - Check the
collectionIdsconfiguration of the knowledge base collection, confirm that it only includes datasets related to new hire onboarding, with no other unrelated business data.
The values given here 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.