DentOP
Client
AI TRANSFORMATION
DentOP AI Transformation
An AI surface for a dental group: a phone receptionist, a self-improving knowledge engine, and a vision model the clinic owns.
Healthcare
Industry
Real-Time TTS, AI Vision
Services
Romania
Region
2025-present
Year
Expert time, spent on repetition.
DentOP runs thousands of inbound calls, WhatsApp X-rays, and treatment plans every month. The call centre burned hours on FAQ calls while real cases waited, and after hours, calls went to voicemail and conversion suffered. Every missed call from an international patient, and DentOP serves diaspora patients from Germany, Italy, Spain, and the UK, was lost revenue.
The clinical bottleneck was the same shape everywhere: expert time spent on repetitive work. Senior operators explaining the same things on every call. Senior doctors reading the same patterns on hundreds of panoramic X-rays a month. The brief was to put AI across the clinic without ripping out the telephony, CRM, and channels the team already runs on.
Call volume
FAQ calls burned the hours real cases needed
Lost revenue
After-hours calls lost to voicemail
International
Diaspora patients across four countries
Repetitive
Doctors reading the same X-ray patterns
Three connected systems, voice, knowledge, and vision.
Three connected systems, built in commercial-logic order. A voice agent on OpenAI Realtime answers the clinic's real phone number over SIP, reasons about caller intent, books appointments, and hands off cleanly to a human, in natural Romanian. It plugs into the clinic's existing telephony (While1) and CRM (AmoCRM), so there is no parallel tooling and no double entry.
Behind the voice agent runs a knowledge pipeline that turns thousands of real call recordings into the agent's retrieval store, with a multi-pass LLM redaction loop that guarantees zero patient PII ever reaches OpenAI. And on the clinical side, a dual-model vision system reads dental X-rays: Claude constrained by an explicit clinical skill, alongside an open-weights model fine-tuned on the clinic's own corpus, so DentOP owns the core intelligence.
Three production systems across the clinic: voice, knowledge, vision.
01.
AI phone receptionist
Voice Agent
An AI phone receptionist on OpenAI Realtime over SIP. A formal call state machine makes the call lifecycle predictable, handoff to a human is clean, and caller memory persists across sessions in Postgres. It plugs into While1 telephony and pushes every transcript and summary into AmoCRM.
02.
The knowledge engine
Knowledge Pipeline
A four-stage pipeline turns thousands of call recordings into the agent's knowledge store. Postgres is the queue; four workers run with exactly-once semantics. A three-pass redaction loop with a two-model gate guarantees zero patient PII reaches OpenAI.
03.
Dual-model X-ray triage
Radiografii
Two models read every X-ray: Claude with an explicit clinical skill, and Qwen3-VL fine-tuned on DentOP's corpus, now 4,900+ annotated cases. Doctors react YES or NO; every NO trains the system. The fine-tuned weights are DentOP's to keep.
The numbers speak.
24/7
Inbound call handling, no added staff
Zero
Patient PII in the knowledge store
4,900+
Annotated X-ray cases, and growing
4
Diaspora markets served in language
The voice agent handles inbound around the clock and improves weekly from the knowledge pipeline. The vision system gets sharper with every doctor correction. The clinic owns the fine-tuned model weights.
Want this across your business?
We find where AI moves your numbers, build the agents and tools to do it, and put them into daily operations. The same work that now runs across DentOP's clinic, 24/7.
Starts with a free 30-minute discovery call.
Next case study
AI copilot for Romanian public procurement evaluation.