Medicall AI agent
Industry
Healthcare
Client
Medical centre
Team
PILOT
Year
2025
We were hired to create an AI assistant that handles overwhelming call volume while maintaining the personal touch our patients expect — and does it all in Lithuanian.
Challenge
The client processes over 95,000 patient calls monthly across 34 operators, causing extensive waiting times and delayed responses due to overwhelming call volume and complex patient inquiries. Manual appointment scheduling and call documentation created inefficiencies that impacted both patient experience and operator productivity. The clinic needed real-time assistance to standardize handling and improve first-call resolution without removing the human touch.
Approach
Built “Medicall,” an AI assistant that transcribes live calls in Lithuanian and suggests context-aware prompts/answers to guide registration and triage in real time.
Deployed bespoke ASR using Whisper + Wav2Vec2, specifically fine-tuned for Lithuanian language speech recognition, achieving minimum 85% transcription accuracy even in noisy call center environments
Integrated with booking and EHR platforms so operators can schedule appointments and view necessary data from a single interface; added quality alerts during the call.
Automated post-call summaries for operator review/finalization to speed documentation and handoffs.
Architecture/Backend
Python, real-time STT, data pipeline for managing live patient interactions.
AI/ML models
ASR Whisper & Wav2Vec2 for LT, optimized for noisy environments (target ≥85% accuracy).
Infrastructure/Deployment
Integration to patient booking platforms, electronic health records.
Key results
Delivered a first-of-its-kind (Lithuania) healthcare call-center assistant prototype, live at KMP, augmenting operators with real-time ASR, guidance, and unified workflows.
Standardized call handling and reduced tool-switching by consolidating five systems into one operator surface, with quality signals during calls.
Designed for scale to other clinics and low-resource languages beyond Lithuanian.
Enhanced appointment scheduling accuracy through AI-powered cross-referencing of doctor availability and patient preferences, reducing scheduling conflicts and errors.
