Frequently asked
questions

MINA is the clinical AI model we built for Turkish clinical practice and the shared core of Opinion AI's four platforms. An open-weight foundation model was adapted to Turkish and to the clinical language of 14 specialties through continued pre-training (CPT); specialty, institution and task expertise was added through LoRA adapters. Knowledge retrieval runs over a clinical knowledge graph (GraphRAG), with agentic orchestration on top of that graph. The user always makes the decision; MINA supports it.

No. MINA was built on an open-weight foundation model that had already learned general language. That model was adapted to Turkish and to the clinical language of 14 specialties through continued pre-training (CPT), and specialty, institution and task expertise was added through LoRA adapters. Knowledge retrieval runs through GraphRAG, and multi-step clinical tasks are carried out through agentic orchestration. We describe the development stages on the MINA page.

MINA was adapted to Turkish clinical language through continued pre-training, not through fine-tuning alone. The open-weight foundation model first went through continued pre-training (CPT), which built Turkish and the clinical language of 14 specialties into its weights. Next, specialty, institution and task expertise was added through LoRA adapters; this expertise is written to separate adapters, not to the foundation model's weights. Finally, GraphRAG added access to the clinical knowledge graph, with agentic orchestration on top of it. This separation keeps language mastery durable in the foundation model and the expertise layer modular.

MINA's clinical knowledge draws on 30M+ sources across 14 specialties: PubMed, The Lancet, JAMA, BMJ, UK Biobank, MIMIC-IV and ClinicalTrials. These sources are linked in a clinical knowledge graph (GraphRAG), and MINA shows the source behind each answer. Institutions' patient data is not added to these sources; institutional data is never used to train MINA's shared model and stays within the institution.

Institutions' patient data is not added to MINA's knowledge base. Institution-specific LoRA adapters are trained inside the institution's own environment on its own data and stay there; institutional data is never used to train MINA's shared model. Continued pre-training of the shared model used Turkish medical terminology, clinical guidelines and anonymized clinical records, meaning records from which no individual can be identified. Opinion AI does not take institutions' patient data into its own systems. MINA runs on the institution's own servers or in an isolated cloud dedicated to the institution, and under KVKK the role of data controller stays with the institution.

We track developments in the medical literature and update MINA's knowledge base. When MINA answers, it uses GraphRAG to find the relevant source in this knowledge base and grounds its answer in it, so every answer can be traced back to its source. Model updates are measured with the evaluation protocols and approved by the Physician Ethics and Advisory Board.

Yes. The architecture we used to adapt MINA to Turkish can be applied to other languages and health systems: continued pre-training adapts the foundation model to clinical text written in the new language, and institution- and task-specific LoRA adapters specialize it for each institution's needs. The POCs we have started with a number of hospital groups in Northern Europe build on this architecture. There too, data stays in each institution's own environment, and MINA runs in a closed environment set up within the GDPR framework.

Opinion AI has four platforms, and all four use the same clinical intelligence, MINA.

TIS (clinical decisions) runs on the hospital and physician side. It covers pre-authorization, the patient digital twin, the clinical assistant (ICD-10, SUT, discharge summaries), the audit cockpit that tracks BH and ADSH indicators, inventory and MKYS tracking, and medical accounting.

SIT-I (insurance integrity) provides control before payment on the insurer side. It works across pre-authorization, policy issuance, claims, reimbursement and medical advisory (Doctor Assistant, Council Mode).

SIT-H (hospital compliance) and SIT-G (public oversight) also run on the same MINA.

On every platform, data stays within the institution and the user makes the decision; MINA supports it.

Both run on the same MINA. TIS (clinical decisions) is built for the hospital operator and physician; it covers pre-authorization, the patient digital twin, ICD-10, discharge summaries, MKYS inventory and audit indicators. SIT-I (insurance integrity) is built for the insurance specialist and manager; it offers file assessment, SUT compliance, policy coverage checks, fraud risk screening and portfolio-level actuarial audit. Both platforms look at the same case with a shared evaluation standard, but data remains within each institution and the two sides' data is never merged.

MINA is the clinical decision-support agent that works alongside the user (ICD-10, SUT, discharge summaries, file assessment and more). MINA Cockpit is the management and portfolio layer of the same MINA. Hospital managers use its audit cockpit to monitor BH and ADSH indicators continuously, while insurance managers see an actuarial portfolio view and above-threshold alerts.

MINA does not make decisions; it produces suggestions. The AI shows the rationale and source behind each suggestion, and every output can be audited. The user always makes every critical decision, such as an approval, a request for missing documents, a rejection or a prescription. Every answer keeps documents and inference apart; it shows clearly which information comes from the patient's documents and which is MINA's inference. Rule checks run in the rule engine; the AI only explains. The GraphRAG architecture ties answers to reference sources, and the source text and confidence score of each prediction are visible.

Opinion AI does not take institutions' patient or clinical data into its own systems or store it. All data remains on the institution's own servers or in an isolated cloud environment dedicated to it, under its control; no data flows to the Opinion AI environment. Under KVKK and the GDPR, the institution is the data controller; if installation, maintenance or support requires access to data, Opinion AI acts as a data processor on the institution's instructions and under a data processing agreement. MINA's computations run on the institution's infrastructure, either on its own servers (on-premise) or in an isolated cloud dedicated to the institution, and personal data is masked inside the institution in line with KVKK requirements. All connections use HTTPS, and storage is encrypted with AES-256. Deployment follows KVKK and GDPR requirements, and Opinion AI holds ISO/IEC 27001, ISO/IEC 27701, ISO 9001, ISO 14001 and ISO 45001 certifications.

No. The HBYS stays as it is; TIS is an intelligent layer that runs on top of your existing HBYS. It reads data from the HBYS and does not write new data back to it, so the integration does not disrupt existing operations. It connects through an XML/CSV export or a direct database connection.

Deployment starts with a single module and expands module by module as needed. The HBYS is not replaced; TIS is added on top of the existing system as an intelligent layer. The timeline is planned together with the institution, based on its infrastructure and the modules selected.

No. The first phase focuses on operators and the finance team. Physicians start using MINA for discharge summaries, ICD-10 suggestions and clinical decision support whenever they choose, within their own workflows. The physician always makes the decision.

Because the system works on a document and process basis, it is compatible with the processes of private health insurers and SGK. Rule sets specific to each insurer can be defined.

Yes. SIT-I also works on its own. Even if the hospital sends documents as PDFs or by email, MINA reads them and produces the same synthesis. If the hospital also uses TIS, both institutions work with the same clinical intelligence and the same rule base; the deployments remain separate, and each institution keeps its own data in-house.

No. MINA only produces suggestions; the specialist always makes the decision to reject, approve or request missing documents. Rule checks run in the rule engine; the AI only explains. The rationale and references (clinical guidelines, SUT clauses, policy terms) are presented to the specialist as decision support.

Yes. Benefit scope for each policy type, waiting periods, contracted institution lists and custom rule sets are defined specifically for your company. The rule engine applies these rules automatically in every file assessment, and MINA explains the result and its rationale.

Opinion AI does not sit between the two sides; TIS runs inside the hospital and SIT-I inside the insurer as separate deployments. Only the data specific to a given application is made available, and only to the specialist who can assess that application. Role-based access control designed around KVKK requirements is applied, so each user sees only the data their role requires. The insurer cannot access other patient files held in the hospital's TIS.

MINA's "Fraud Risk Analysis" query works at the portfolio level as well as on individual files. Recurring diagnosis patterns, hospital–physician pairings, atypical procedure–policy relationships and duplicate billing signals are scanned systematically. Suspicious patterns appear in MINA Cockpit's daily report as above-threshold alerts.

In the contact form on the home page, choose “POC Request” under “Area of Interest” and briefly describe your institution and the process you want to evaluate. Our team defines the scope of the POC (proof of concept) together with your institution.

We share the architecture, evaluation method and deployment details with your institution's technical team during the POC meeting. You can send your POC request through the contact form.