Multi-Turn Conversation and Prompts for University Laboratory Safety Classification: Handle Questions and Collect Missing Information

Lab Safety Categories: review conversation for question intake. Check source versions, input mappings and the values received by downstream nodes.

What this category's data looks like

In the university laboratory safety grading and classification scenario, intake materials usually include the laboratory grading and classification table, the hazard source list, and the safety responsibility statement. The grading and classification table registers the category and risk level by room or laboratory unit, the hazard source list records items such as chemicals, gas cylinders, and equipment, and the safety responsibility statement carries the responsible person and the signed receipt. The three types of materials come from different sources: the grading and classification table is filled in by the laboratory and updated as the rating is adjusted, the hazard source list changes with experimental projects and reagent stock in/out, and the responsibility statement is signed by academic year or upon personnel changes. Field characteristics center on name, quantity, date, and signature: the same room may be written differently in the two materials, quantities and dates are prone to misalignment, and when versions are not synchronized or the signed receipt is missing, intake cannot proceed directly. After the materials enter the Q&A intake step, the grading and classification table is used to confirm unit attribution and level, the hazard source list is used to check items and quantities, and the responsibility statement is used to confirm whether the signature and receipt are complete; missing items trigger follow-up questions.

Configuration and implementation checks

This step is executed by the memory turn count and prompt configuration in the app settings, which output the referenced historical message range and answer constraints. The intake record and the completed follow-up questions are then handled by the model. The memory turn count controls how many previous turns of dialogue can be referenced at most when answering; if the value is too small, the preceding text cannot be used, and if it is too large, irrelevant content is brought in and consumption increases.

ObjectCheckVerification action
Laboratory grading and classification tableCheck the name, quantity, date, and signature of the same record against the hazard source list and safety responsibility statement; if versions are not synchronized or the signed receipt is missing, it is judged as a missing itemCheck the memory turn count value in the app settings, and run a trial with two consecutive turns of questions to observe whether the second turn carries over the reference from the first turn
Hazard source listThe item name, quantity, and date must match the corresponding unit in the grading and classification table; if the same record does not match across the two materials, it is judged as a missing itemWrite the historical message range and answer constraints into the prompt, and check whether the preceding text in messages is correctly referenced
Safety responsibility statementThe signature and receipt fields must be complete, and the date and personnel change records must be synchronized with the grading and classification table; if the signature or receipt is missing, it is judged as a missing itemUse chatId to distinguish sessions, check whether variables are updated with the follow-up questions, and avoid treating the opening message as preceding text
Intake record and completed follow-up questionsMissing items are identified and followed up on, answers are based on materials, and after completion they are forwarded to the handlerCompare the usage records of two runs to confirm that consumption changes accordingly after the turn count is adjusted

Three common errors

  • Symptom: the second question uses a reference and the answer is off-topic. Cause: the memory turn count value is too small, the preceding information is not included in the reference range, and the model cannot carry over the object from the previous turn.
  • Symptom: the name, quantity, and date of the same record do not match across the two materials. Cause: the versions of the grading and classification table and the hazard source list are not synchronized, and they were not updated in sync after filling in.
  • Symptom: the responsibility statement is still accepted when the signature or receipt is missing. Cause: the prompt does not constrain the check of the signature and receipt fields, and the missing item is not identified and followed up on.

How to verify the setup

This feature acceptance: ask two consecutive follow-up questions and see whether the second question carries over the reference from the first question. Give a certain laboratory unit in the first turn, ask with a reference in the second turn, and check whether the answer continues to use the object from the first turn without asking again.

Business test: ask a question that lacks key information and see whether it first follows up before answering. Give only the room number without the category or hazard source item, and observe whether it first asks for the name, quantity, date, and signature.

  • Check the current memory turn count value in the app settings and match it with the preceding text range required by the question; run a trial with two consecutive turns of questions to confirm that the second turn can correctly carry over the reference from the first turn.
  • Compare the usage records of two runs to confirm that usage changes accordingly after the turn count is adjusted; check whether the answer constraints in the prompt cover the check of the signature and receipt fields.