Overview
Sirma is redesigning the front end of clinical care by embedding agentic AI into the patient intake process, ensuring physicians spend less time on administrative data collection and more time on diagnosis and treatment. Utilizing the capabilities of Sirma.AI and Medrec:M Clinic, Sirma delivers an enterprise-grade AI agent workflow that captures a patient’s history, symptoms, and risk factors before the consultation even begins, preparing a structured clinical summary that the physician can verify, edit, or reject.
The Challenge
Healthcare systems throughout Europe are facing growing financial and operational challenges, with Bulgaria particularly acute. From 2005 to 2012, total health spending in Bulgaria grew by 13.5% annually, outpacing GDP growth. Between 2009 and 2019, per-capita health expenditure increased by 83%, compared to just 28% across the European Union. As of 2023, Bulgaria is still spending only about half of the EU average per person on healthcare, while out-of-pocket expenses remain the highest in the Union.
In addition to funding issues, another critical concern is clinician time, the most limited resource in medicine. A significant portion of this time is spent on tasks that are only partially clinical, such as collecting patient histories, reconciling documents, and converting patient information into digital data.
The Sirma Approach
One of the critical process improvements that the Sirma solution offers is a redesigned pre-consultation workflow built around agentic AI rather than passive digital record-keeping. Within defined clinical boundaries, the system:
- Interviews patients before their visit, capturing problems, symptoms, family history, medications, allergies, surgeries, and risk factors
- Translates patient language into structured clinical information and flags urgent symptoms for review
- Prepares concise clinical summaries that physicians can verify, edit, or override before the consultation starts
- Orchestrates triage, documentation, and next-step routing without replacing clinical judgment
- Operates as an embedded service layer connecting patient records, telehealth, scheduling, and physician workflows, rather than a standalone chatbot
This architecture positions Sirma.ai and Medrec:M Clinic not as add-on tools but as an integrated clinical software layer, ensuring AI outputs remain 100% traceable to source documents and easy for clinicians to audit.
Governance and Trust
Sirma treats human oversight as a design principle rather than a compliance checkbox. Higher-autonomy AI is applied to low-risk, repetitive tasks such as intake and documentation, while human accountability rises alongside clinical complexity. This approach aligns with the EU AI Act, which classifies many medical-purpose AI systems as high risk and requires risk mitigation, data quality, transparency, and human oversight. Audit trails, consent management, privacy protection, and continuous monitoring underpin every deployment.
Results and Impact
Deployment priorities focus on narrow, high-volume workflows including pre-consultation intake, follow-up summaries, chronic disease monitoring, referrals, and patient communications, each measured against concrete outcomes:
- Reduced administrative time per clinical encounter
- Shorter overall patient journeys and fewer repeated explanations
- Faster escalation of risk and higher-quality documentation
- Improved clinician satisfaction and fewer unnecessary visits
- Greater patient understanding of, and control over, their own care
Technologies
The pre-consultation intake capability in Sirma.AI and Medrec:M Clinic is generated using a combination of supervised machine learning algorithms and large language models (LLMs) tuned specifically for clinical language understanding and structured data extraction.
-
Advanced Speech-to-Text and Conversational AI Models: Deployed on the Sirma.AI platform, these models accurately capture and transcribe patient responses during the AI-led intake interview, whether conducted via voice or text.
-
Natural Language Processing (NLP): The AI framework identifies and interprets medical entities — symptoms, complaints, medications, allergies, family history, and risk factors — from unstructured patient narrative.
-
Natural Language Generation (NLG): GTP-driven generation produces concise, contextually accurate clinical summaries structured to meet physician documentation standards, ready for the doctor to verify, edit, or reject.
-
Secure Cloud Infrastructure: HIPAA compliant and available sovereign deployment environments ensure data privacy, consent management, and regulatory alignment throughout intake, storage, and processing, consistent with EU AI Act requirements for high-risk medical AI.
-
Integrations: Sirma integrates direct connectivity with electronic health record (EHR) systems, telehealth platforms, and physician scheduling tools, which enables seamless user experiences.
-
Cost-Efficient, Scalable Architecture: System design balances performance with affordability, supporting deployment across high-volume outpatient settings without adding hardware or staffing overhead.
Key supervised AI algorithms powering the intake workflow include:
-
Transformer-based Models: Fine-tuned variants of BART, BioBERT, and related bidirectional transformer architectures convert free-form patient statements into accurate, structured clinical entries.
-
Sequence-to-Sequence Models: The Sirma.AI deployed agents translate patient narrative into summarized clinical intake notes while preserving relevant symptoms, context, and reported risk factors.
-
Named Entity Recognition (NER) and Relation Extraction (RE): Custom AI agents designed to identify clinical symptoms, medications, allergies, surgical history, and their relationships, enriching the intake record with structured medical information.
-
Section-wise Model Training with Adapters: Modular training tailored to distinct intake domains (history of present illness, medications, family history, risk factors) improves precision and allows flexible summary composition.
-
Fine-tuned Large Language Models (LLMs): Sirma has the expertise and capability to create domain-adapted models generate structured clinical summaries, aligned to formats physicians already use and integrating medical knowledge for high-fidelity, review-ready output.
-
Risk-Flagging and Triage Logic: A dedicated virtual teal of Ai agents acts as a classification layer and scans structured intake data in real time to flag urgent symptoms for immediate physician attention, supporting the human-in-the-loop principle central to Sirma’s clinical AI design.
This technology stack enables Sirma.AI and Medrec:M Clinic to transform unstructured, pre-visit patient conversations into structured, physician-ready clinical summaries, reducing documentation time before the consultation even begins while preserving full traceability and clinician oversight.
Why It Matters Now
For countries facing ageing populations, clinician shortages, and rising burdens of chronic disease, this is a capacity strategy, not a luxury. Every minute recovered from avoidable paperwork can be reinvested into access, quality, and empathy. Bulgaria, and markets like it, are well-positioned to move early because the pain points are visible and the underlying digital infrastructure already exists.
Sirma’s platforms offer a pragmatic, near-term path to a system that finally treats clinician attention as the precious resource it is, protecting the doctor’s time, strengthening the patient relationship, and making care delivery measurably more efficient.
