Thalion provides a self‑hosted, GDPR‑compliant large language model fine‑tuned on biomedical instruction data, running on dedicated EU data‑center infrastructure. It offers a real‑time chat interface and a documented REST API for secure integration with electronic health record systems and clinical workflows, featuring low‑latency token streaming, end‑to‑end encryption, and ISO/IEC 27001‑certified hosting.
Funding
Funding not disclosed
Founders
Product
Problem
Healthcare providers need AI-driven assistance for diagnostics, documentation, and clinical decision support, but existing large language model services often rely on third‑party cloud APIs that expose patient data to non‑EU jurisdictions and raise GDPR compliance concerns.
Solution
Thalion delivers a self‑hosted, GDPR‑compliant large language model specifically fine‑tuned on biomedical instruction data (UltraMedical‑8B, a Llama‑3‑8B variant). The platform runs entirely on dedicated EU‑based servers, ensuring that all user prompts and generated outputs remain within the European data residency boundary. Users interact via a real‑time chat UI or integrate the model through a documented REST API, enabling seamless embedding into electronic health record (EHR) systems, clinical workflow tools, or custom applications. The service combines low‑latency token streaming with parallel llama.cpp workers to handle concurrent requests while maintaining high availability. Security controls include TLS encryption in transit, at‑rest encryption where applicable, role‑based access, and ISO/IEC 27001‑certified hosting, providing a trustworthy AI assistant for medical teams.
Target Audience
Primary customers are hospitals, clinic networks, and health‑tech startups that require secure, domain‑specific AI assistance for clinical documentation, diagnostic support, and workflow automation.
Features
- UltraMedical‑8B: Llama‑3‑8B model fine‑tuned on ~410 k biomedical instructions and >100 k preference pairs using supervised fine‑tuning and preference learning (DPO/KTO)
- Self‑hosted deployment on dedicated EU data‑center infrastructure, eliminating reliance on external cloud AI providers
- Real‑time chat interface with SSE‑based token streaming and a bounded FIFO queue for low‑latency, high‑throughput request handling
- Fully documented REST API supporting programmatic access, custom workflow automation, and integration with existing health IT stacks (EHR, RIS, PACS)
- End‑to‑end encryption (TLS/HTTPS) and optional at‑rest encryption, with role‑based access controls and audit logging
- ISO/IEC 27001‑certified hosting environment, ensuring compliance with GDPR and other European data‑protection standards
- Scalable licensing model that includes per‑user seats, team collaboration features, and priority support options