Document checks for lab and quality teams
Certificates of analysis, batch records, sample intake and deviations. Agents flag what is out of specification; qualified people make every release decision.
The playbook in 54 seconds
What each agent checks and who decides the exceptions. All videos · Also on YouTube
What you can build for virology and life sciences
Certificate of analysis review
Certificate of analysis review reads a batch certificate and your specification sheet and pulls out each test result and limit. It checks every result against its specification limit and that the certificate is signed and dated, then sends anything out of specification to quality assurance with a review summary.
Tested: clean set completed; 3 of 3 planted errors flagged.
View on the Docy AI marketplace →See the brief
Build an agent that reviews a certificate of analysis for a viral product batch. It has two inputs: the certificate and the product specification sheet. From the certificate, extract the product name, batch number, test date, signature and each test result with its unit (titre, purity, endotoxin, sterility, residual DNA, mycoplasma). From the specification sheet, extract the limit for each test. Check each result against its specification limit, and check that the certificate is signed and dated. If any result is out of specification or the certificate is incomplete, send it to quality assurance for a decision. Then write a short review summary listing each result, its limit and the outcome.
Batch record review
Batch record review reads a manufacturing batch record. It checks every process step is completed, signed and dated, and every deviation has a reference number, then sends gaps to the quality reviewer with a short summary.
Tested: clean set completed; 3 of 3 planted errors flagged.
View on the Docy AI marketplace →See the brief
Build an agent that reviews a manufacturing batch record. Extract the batch number, product, each process step with its operator, date and signature, and any recorded deviations. Check that every step is completed, every step is signed and dated, and any deviation has a reference number. If anything is missing, send it to the quality reviewer for a decision. Then write a short batch review summary listing the gaps and the outcome.
Sample intake
Sample intake reads the submission form and chain-of-custody document. It checks required fields, storage temperature and signatures at every handover, and sends problems to the lab manager to accept, reject or ask for more information.
Tested: clean set completed; 3 of 3 planted errors flagged.
View on the Docy AI marketplace →See the brief
Build an agent that checks incoming laboratory samples. From the sample submission form and chain-of-custody document, extract the submitter, sample type, collection date, storage temperature, number of samples and handling requirements. Check that all required fields are filled, the storage temperature is within the allowed range of 2 to 8 degrees C, and the chain of custody is signed at every handover. If any check fails, send it to the lab manager to accept, reject or request more information. Then write a short intake record.
Deviation and CAPA review
Deviation and CAPA review reads a deviation report. It checks a root cause is stated, every corrective action has an owner and due date, no open action is past the report's review date and quality has approved, then sends gaps to the quality manager with a summary.
Tested: clean set completed; 3 of 3 planted errors flagged.
View on the Docy AI marketplace →See the brief
Build an agent that reviews deviation and corrective action reports. Extract the deviation number, date, review date, description, root cause, corrective actions with owners, due dates and status, and approvals. Check that a root cause is stated, every corrective action has an owner and a due date, no open action has a due date earlier than the review date on the report, and the report is approved by quality. If any check fails, send it to the quality manager for a decision. Then write a short quality summary of the findings.
The brief is what you type into the Docy AI assistant. Agents tagged Open Docy can also be built by hand on the open-source canvas. Agents tagged Docy AI may need custom code, a knowledge list or a policy node, which Open Docy does not include.
Your rules. A person's decision. A record.
Send the records
Upload certificates, batch records, intake forms and deviation reports.
Apply your specifications
Your limits and procedures are checked the same way every time. Anything out of specification or incomplete stops and goes to a named person.
Keep the record
Each run records the result, its limit and who made the decision.
Be clear about the limits
- No scientific or clinical judgment. Release, safety and diagnosis decisions stay with qualified people.
- No regulatory validation. Regulated manufacturing needs computer-system validation before use; that work is the customer's.
- Sensitive data may call for an on-premise or private deployment instead of a shared cloud.
- These agents check documents against rules you set. They do not replace the people who make the decision.
- We have tested these: each completed a clean test set and flagged the errors we planted in a flawed one. Test on your own files before relying on them.
Want to try one on your own virology and life sciences documents?
Tell us which documents you work with and which check would save you the most time.