Batch-record review of 1,547 lot dossiers per year collapsed from 6 hours to 38 minutes
A pharma QA agent reads each lot dossier, validates fields against the MBR and SOP, cites the page and excerpt of every exception, and routes only the flagged dossiers for human release. A second agent attacks R$1.13B of stagnant warehouse inventory and the public-tendering pipeline.
120-year-old public vaccine and antivenom institute. Naming withheld.
The institute manufactures vaccines and antivenoms on behalf of a public health system, holds 120 years of process history, and runs its operations on SAP. Two distinct workflows were bleeding capacity. Lot dossier review consumed six hours per batch across 1,547 batches per year. And the warehouse held over a billion reais of stagnant inventory the procurement team could not move because the public-tendering process for new buys kept stalling.
Outcomes
Two of the institute's heaviest manual workflows ship with the same SAP, the same data boundary, and a fraction of the human hours.
The problem
Every lot dossier ran six hours of manual review: open the document, check fields against the master batch record, verify signatures and timestamps, cross-reference contract clauses and regulatory adherence, and sign off. Parallel to that, the warehouse held an R$1.13B inventory the institute could neither use nor unblock, while procurement waited on licitação cycles that ran on a calendar of their own. Both problems sat on top of SAP and could not move outside the institute's data boundary.
What we built
HomoDeus deploys two agents inside the institute's SAP environment. The Batch-Record Review agent reads each lot dossier, validates every field against the master batch record and SOP, cites the page and excerpt of every exception, and queues only the flagged dossiers for human release. The Warehouse and Procurement agent reads the SAP inventory, models the addressable portion of the stagnant stock, and runs a buyer-intelligence layer that compares prices and qualifies suppliers under Anvisa rules for the public-tendering pipeline.
Stack
- Batch-record review agent
- MBR/SOP rule matcher
- Page-cited exception engine
- SAP-adjacent integration (no migration)
- Warehouse inventory intelligence agent
- Supplier qualification agent (Anvisa-aware)
- Public-tendering (licitação) workflow
Time to production
Phased: batch-record agent first, procurement agent following
Questions a C-suite leader will ask
Can AI review pharma batch records under GMP?
Yes. The agent validates each field of the lot dossier against the master batch record and SOP, cites the page and excerpt of every exception, and routes only the flagged dossiers for human release. The human decision remains GMP-anchored; the rote validation runs in minutes instead of hours.
Does the system require us to leave SAP?
No. The agents are SAP-adjacent. They read and write against SAP without migration, the institute keeps its system of record, and the data stays inside the institute's environment.
How does an AI procurement agent unlock stagnant warehouse inventory?
It models the addressable portion of the inventory against current operational need and pricing, then runs a buyer-intelligence layer that compares prices and qualifies suppliers under Anvisa rules. The agent does not replace the public-tendering process; it accelerates the analysis that feeds it.