Opinion AI is a real-time clinical and financial decision intelligence platform for hospitals, insurance and public oversight. Our model, MINA, has been adapted to Turkish clinical language. It gives physicians a second opinion, checks payments before they go through and cuts the time teams spend on documentation. The platform runs inside each institution's own environment, and the data never leaves it.
Explore the PlatformMINA runs the same clinical intelligence on four platforms. TIS supports the clinical decisions of physicians and hospitals; SIT-I serves insurers, SIT-H hospitals' compliance processes and SIT-G public oversight. Each platform runs inside its own institution, following that institution's workflow.
When physicians open a patient, they see the history, tests and discharge summaries in one profile, with ICD-10 and SUT checks on the same screen.
For hospitals →It shows the deduction risk and the items to fix before an invoice is submitted, with the rule behind each one and its owner.
Hospital compliance →From pre-authorization to reimbursement, coverage, limits, file history and SUT compliance reach the specialist ready for review.
For insurers →It presents every finding with its basis in SGK invoice review and GÖREN assessment; the authorized auditor makes the decision.
Public oversight →See the patient's clinical history on one screen, with early risk signals and sourced recommendations in front of you at the moment of decision.
ExploreTIS is added as a layer on top of your existing HBYS (hospital information system). Pre-authorization, audit, MKYS inventory and medical accounting are followed in the same cockpit.
ExploreEach pre-authorization request is assessed against current clinical guidelines; missing documents, coding errors and fraud risk are flagged before approval.
ExploreLab results flow into the patient's data twin; batch OCR, automated analysis and SUT compliance checks run in the same flow.
ExploreThe patient twin brings HBYS and Medula records into one live profile; the physician twin tailors recommendations to the physician's own practice.
ExploreData stays inside the institution. Storage is local and set up for KVKK and GDPR requirements; data is encrypted with AES-256, and every output is presented with its rationale and can be audited.
ExploreToday the information about a single patient sits in several systems, in different formats. The data twin brings those pieces together and puts a readable picture in front of the physician.
Records from HBYS and Medula merge into one live patient profile. Discharge summaries and reports written as free text are read and turned into structured fields.
When physicians open the patient, they find the clinical summary, risk flags and SUT checks in the same view.
It learns the physician's test-ordering patterns, coding preferences and treatment patterns. It is created with the physician's consent and compared against anonymized peers.
Recommendations are therefore based on the physician's own practice rather than a generic pattern, and they always come with their rationale.
MINA works inside day-to-day tasks such as pre-authorization, billing, audit and inventory. Teams no longer start every file from scratch. The points below describe what changes in daily work.
Task-level evaluation results for MINA v3.7 are on the MINA v3.7 page.
Operators no longer build each file from scratch; they start from the clinical synthesis and risk score MINA has prepared.
Missing documents and incorrect ICD-10 codes surface before the application reaches the other party, so they can be fixed in time.
BH and ADSH indicators are monitored continuously; risks are flagged before the audit arrives.
MKYS (the Ministry of Health's inventory management system) data is scanned every day against 60- and 90-day thresholds, so silent budget loss becomes visible early.
The platform is added on top of the existing system; the team keeps its familiar workflow.
Pre-authorization, audit, inventory and medical accounting indicators come together in the same cockpit.
The platform runs on the institution's own servers or in an isolated cloud environment. The data layer is reachable only from the local network; connections use HTTPS and storage is encrypted with AES-256. HBYS, MKYS, SGK (social security) and clinical data flows are normalized automatically, and personal data is masked in line with KVKK.
Data is not transferred out. The database is reachable only from the local network; administrators connect through an SSH tunnel.
All connections use HTTPS; data is stored on AES-256-encrypted disks.
It integrates through XML/CSV export or a direct database connection and sits on top of the existing HBYS.
MINA is there to support decisions; every action stays with the user, and every output comes with its rationale and source.
We work inside live operations with insurers and healthcare institutions. Our pre-authorization collaboration with AXA Sigorta and our selection for Allianz Türkiye's HackZone program were covered in the press. At large private hospital groups in Türkiye, we run as an intelligent layer on top of HBYS, and we have started POCs with a number of hospital groups in Northern Europe.
Our clinical decision support system is being integrated into AXA Sigorta's pre-authorization processes. Diagnoses and requested procedures are checked for alignment with current medical guidelines; the final decision always rests with AXA Sigorta's medical specialists.
Read the story →We were selected from nearly a hundred applications to the Turkish insurance sector's first open innovation program and ran a POC with Allianz teams throughout the program. We presented MINA on stage at Demo Day on 16 September 2026.
Read the story →In these hospital groups, TIS runs on top of the existing HBYS without changing it. HBYS and Medula data merge into one live patient profile, and ICD-10 and SUT mismatches are flagged before an invoice goes out; the data stays inside each hospital.
Read the story →We have started POCs with a number of hospital groups in Northern Europe. There too, MINA runs in a closed environment set up within the GDPR framework and the data stays within each institution's own systems.
Read the story →MINA is developed by a 17-person multidisciplinary team, and its clinical accuracy is reviewed by a board of physicians from 15 specialties. Because the quality of a model depends largely on the data it is trained on, data engineering is as central to our work as model development.
Our data flow is built on medallion data layers that run inside the institution's own environment. In the Bronze layer, raw data from the institution's hospital information systems, insurance pre-authorization records and clinical documents is ingested; in the Silver layer it is cleaned and normalized, and personal data is masked in line with KVKK requirements; in the Gold layer, structured datasets ready for clinical analytics, billing compliance checks and digital twin models are produced. The data does not leave the institution, which remains the data controller.
For clinical literature and reference knowledge, we use a dual knowledge graph approach: peer-reviewed articles and clinical guidelines are indexed in a graph database, while the medical textbook corpus is indexed in a vector search store. Routing between these two sources is managed dynamically by an ML routing layer according to the clinical context.
The model layer has three tiers. An open-weight base model is adapted to Turkish and to the clinical language of 14 specialties through continued pre-training (CPT). Specialty, institution and task expertise comes from LoRA adapters. Knowledge retrieval runs on GraphRAG over a clinical knowledge graph, with agentic orchestration working on top of that graph. All training runs on high-performance GPU infrastructure, and all data processing and inference are carried out in line with KVKK requirements.
This architecture is developed and maintained by a team of 17 people, ranging from AI researchers to data engineers, from clinical domain experts to quality assurance and regulation specialists. Every member of the team is an expert in their own field. Our technical core of data scientists, ML engineers and senior AI specialists specializes in data warehouse design, model training, scalable infrastructure and clinical safety engineering.
Alongside the technical team, we have a 15-member Physician Ethics and Advisory Board. Physicians from 15 specialties regularly review the model's clinical accuracy and ethical boundaries. The board brings together specialists with distinguished academic and clinical careers in cardiology, oncology, internal medicine, radiology, emergency medicine, obstetrics & gynecology, pediatrics, neurology, orthopedics, pathology, anesthesiology, general surgery, urology, pulmonology and psychiatry. It regularly reviews MINA's clinical knowledge base, case scenarios and decision support outputs, and signs off on model updates.
People always make the decision. MINA shows the rationale behind its recommendation and the source it relies on, and every output can be audited afterwards. We have also contributed to national policy: we submitted our views during the preparation of the AI Action Plan, and the risk-based regulation, EU alignment, anonymized data sharing and regulatory sandboxes we proposed were included in the 2026 Action Plan.
Our team writes about how the platform works, what it changes in practice and where AI in healthcare stands today.
The patient twin merges HBYS and Medula data into one live profile; the physician twin turns test, coding and treatment patterns into personalized, reasoned decision support. The data stays inside the institution.
Read article →From pre-authorization to reimbursement, policy analysis, limit tracking, file history, SUT compliance and clinical context reach the medical reviewer ready on one page. The final decision stays with the insurer's medical specialists.
Read article →In health and insurance workflows, the rule engine carries out the assessment, AI explains the reasoning and the source, and a person makes the decision. The article covers why document and inference are kept apart, the six separate rules defined for six pharmacy rejection types, and how every output can be audited.
Read article →For a POC request, a deployment question or a partnership, fill out the form; your message goes directly to the Opinion AI team.