hellomed, founded in Berlin in 2022, is a mail-order pharmacy for professional care that supplies patients in more than 100 care locations with ready-to-take blister packs. Gemma Analytics built AI agents for document processing that read, sort and match incoming faxes, scans and emails, so the team only checks the unclear cases.
The starting point
To supply patients, hellomed needs many documents from care facilities, doctors and pharmacies: SEPA direct debit mandates, consent forms, co-payment exemptions, medication plans. They arrive by fax, as scans, as photos or by email, and all landed in one shared inbox. For each document, a person had to read it, recognise the type, assign a team, find the patient, split bundled files, rename the PDF, upload it and type in the data by hand.
This cost time on every document and led to typing errors. Because the content was only known after opening, nothing could be prioritised early, and many tickets ended up in one pot labelled “Other”. A single file could contain up to 50 patient documents, and it was often passed from one team to the next.
What we built
Gemma Analytics built two AI agents that work directly on the inbox. The triage agent reads every incoming document, recognises its type, splits bundled files into individual documents and finds the matching patient in hellomedOS, hellomed's own platform. The document admin agent then extracts the data, for example from SEPA mandates, consent forms and exemptions, and uploads the document to hellomedOS once it has been approved.
People stay in control. Whenever the agent is not sure, for instance when one fax holds documents for two patients or the type is unclear, it hands the case to a person instead of guessing. The team approves every upload and closes every ticket itself. The project started in summer 2026. The triage agent went live in August, the document admin agent followed in September.
The result
Both agents are live in production. hellomed gets 134 hours of working time back every month, and the same team now handles 27% more tickets per working day. 82% of tickets pass every step, from classification and patient matching to data extraction, with high confidence. The team's time now goes to the cases that need a person.
Tools and setup
The agents run as automated workflows in n8n and use AI models from Mistral AI to read and classify documents and extract their data. Zoho Desk stays the team's ticket system: documents arrive there as tickets, the agents add their results, and the team reviews the open cases in the same place. Approved documents land in hellomedOS, so the automation fits into systems the team already uses.