Ivatherm · Case study
A voice CRM for pharmacy visits
Reps dictate pharmacy visits in Romanian. Ivatherm’s CRM drafts the visit record and follow-up tasks, then links each finding in the manager’s report back to the visits behind it.
- Industry
- Dermocosmetics
- Territory
- Romania
- Project
- SOME1 · 2025
The brief
Ivatherm’s field representatives visit pharmacies across Romania. Their managers needed to know what pharmacists were asking for, which products came up and what the rep had promised to do. The details often landed in a spreadsheet hours after the conversation.
The team wanted a mobile CRM a rep could operate with one thumb, straight after a visit. A manager needed to see activity across the territory and ask a question about what had happened. Work, Meet AI’s founder delivered the original project at SOME1.
From voice to visit record
The rep opens a visit, taps record and speaks in Romanian. The voice agent transcribes the account in real time. Questions can change with the type of customer, so the rep hears prompts relevant to that meeting.
The CRM extracts the products discussed, the pharmacist’s reaction and any next steps. The record stays attached to the pharmacy. A manager can ask a question in everyday language, read a report across visit history, and open the map to see where those conversations took place.
Try a sample visitThe territory on screen
Clients and pharmacy chains have geocoded locations. The live map places accounts and visits together so a manager can move between the national picture and a particular stop.
What shipped
- Mobile CRM
- A phone-first workspace with separate representative and manager roles.
- Spoken visit notes
- A Romanian-language capture flow that organises a spoken account into CRM fields.
- Reporting
- A conversational interface for business intelligence across the visit history.
- Account directory
- Client and pharmacy-chain records with geocoded addresses.
- Territory view
- A live map displaying clients and visits.
- Question design
- Administrators can adapt the voice agent’s questions to each type of client.
- Report exports
- Managers can take reports into the next meeting as Excel files or PDFs.
Inside the implementation
In the original 2025 implementation documented by SOME1, ElevenLabs Conversational AI handled the live Romanian conversation and transcription. GPT-4o extracted the visit fields, with checks for details unsupported by the conversation. GPT-5 answered managers’ reporting questions using the recorded visits.
Supabase Postgres enforces row-level access by role. Google Maps provides address autocomplete and geocoding, and Mapbox GL renders the territory. The application passes the administrator’s client-specific prompts to the voice agent.
- Next.js 14
- TypeScript
- Supabase Postgres + RLS
- ElevenLabs Conversational AI
- OpenAI GPT-4o / GPT-5
- Google Maps
- Mapbox GL
- shadcn/ui
- Tailwind CSS
- Vercel
Project credits
Eduard Cristea worked on Ivatherm CRM at SOME1. The original public project documents the CRM scope, stack and territory film. View the original Ivatherm case at SOME1.

