What the data for this category looks like
Typical entities are skincare, haircare, makeup, and special cosmetics manufacturers. Daily documents include formula work sheets, weighing and feeding records, emulsification or filling records, cleaning and sanitization records, semi-finished and finished product inspection reports, and label review documents. Data sources include registered or filed formulas, raw material safety materials, weighing and emulsification equipment data, ERP/MES systems, laboratory test results, and packaging material version documents. Data is updated every production batch or each line change. Document fields include raw material name, restricted substance content, emulsification temperature, batch number, packaging material version number, and others. Units include degrees Celsius, grams, milligrams, and others.
Constraints imposed on knowledge base retrieval and recall
Bid preparation and quotation creation for cosmetics manufacturing rely on compliant filed data and production records. The high-frequency update feature requires the retrieval system to support real-time or near-real-time document synchronization, to avoid using expired formulas or label content. The multi-type, finely segmented document structure requires retrieval to accurately match business fields such as restricted substance content and packaging material version number, to avoid generalized matching that leads to compliance risks. In bid scenarios, documents matching the current project must be located quickly, so the recall scope must be limited to the filing library and production record library of the corresponding category, to avoid irrelevant documents interfering with the review process.
Configuration settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Filter low-match non-compliant documents to avoid using unfiled formulas or label content during bids |
limit | Top 10–15 results | Control the number of recall results, reduce the workload of bid document review, while covering core compliance basis |
searchMode | mixedRecall | Match both formula text and precise filed fields such as raw material names and restricted substance content, improving retrieval accuracy |
usingReRank | Enabled (re-rank window for top 20 results) | Re-sort recall results by the category of the bid project (such as makeup, skincare), prioritizing documents matching the current project |
datasetSearchUsingExtensionQuery | Enabled | Automatically expand query terms, for example, expand "formula" to "filed formula, work sheet", covering all compliant document types required for bids |
collectionIds | Bind 5–8 specified document collections | Limit the retrieval scope, avoid recalling documents from irrelevant categories, improving retrieval efficiency and compliance |
Three common misconfigurations
- The issue is that retrieval results include packaging material version documents not related to the current bid project. The cause is that
collectionIdsis not configured to limit the retrieval scope, and the full knowledge base content is mistakenly included in the recall. - The issue is that the latest filed formula from the knowledge base is not retrieved. The cause is that automatic synchronization trigger rules are not configured, and only manual updates to the knowledge base collection are performed, so the latest production or line change data cannot be obtained.
- The issue is that the retrieval request returns a 429 status code or takes more than 30 seconds. The cause is that the retrieval frequency configuration is not adjusted according to the project, exceeding the call limit of the current knowledge base, and the request frequency of single retrieval is not reasonably limited.
How to confirm the configuration is correct
- Enter the filing formula number of the bid project, check whether the corresponding formula work sheet and raw material safety materials are returned first in the retrieval results.
- View the retrieval log, confirm that only the document collections corresponding to the currently bound
collectionIdsare included, with no documents from irrelevant categories. - Test whether after enabling
usingReRank, the results for makeup bid projects preferentially return makeup-related inspection records and packaging material documents. - Enter the query term "label", confirm that the retrieval results include label review documents and filed claim content, verifying that
datasetSearchUsingExtensionQueryis functional.
The values provided are common starting points and should be measured against the reader's own samples.