The Challenge
A healthcare organization needed a patient intake chatbot to streamline front-desk operations and improve the patient experience. However, they faced significant obstacles with their initial GPT-3.5 approach.
- Prohibitive usage fees — High-volume patient interactions made API costs unsustainable at scale.
- Data sensitivity — Patient health information demanded on-premise infrastructure for regulatory compliance and data governance.
The organization needed a solution that delivered comparable conversational quality at a fraction of the cost, while keeping all patient data within their own infrastructure.
The Solution
Comerit deployed a purpose-built AI system tailored to the healthcare organization's specific requirements.
- Fine-tuned Mistral-7B — A customized open-source model trained specifically for healthcare patient intake operations, delivering results comparable to GPT-3.5 at dramatically reduced expense.
- On-premise infrastructure — Complete data governance and privacy compliance through self-hosted deployment, eliminating third-party data exposure.
- Comprehensive quality assessment — Performance evaluated across four key dimensions:
- Accuracy in query interpretation
- Fluency of communication
- Comprehension of medical terminology
- Empathy in patient interactions
The Results
The fine-tuned Mistral-7B deployment exceeded expectations across every measured dimension.
- Performance parity — Achieved results on par with GPT-3.5 across all key metrics.
- Consistent quality — High-quality patient interactions maintained across thousands of daily conversations.
- Complete data privacy — All patient data remained within the organization's on-premise infrastructure.
- Substantial cost reduction — Operational costs reduced significantly compared to the original cloud-based approach.
This engagement demonstrates that open-source models, when properly fine-tuned and deployed, can match proprietary alternatives while providing superior cost control and data governance.


%2520(1).png&w=3840&q=75)