Advanced Learning Medical Assistant ALMA Project
Conversational AI as a Catalyst in Modern Healthcare

In today's healthcare sector, information overload and the need for quick and informed decisions represent a daily challenge for physicians and healthcare professionals.
Conversational artificial intelligence is emerging as a transformative solution to assist in this environment: models such as OpenAI's GPT-4 or Anthropic's Claude have demonstrated the potential of AI to entangle conversations and answer complex questions. However, generic versions of these models are not always suited to the demands of the clinical setting. They can provide accurate information, but sometimes make significant errors or omissions, or fail to present answers with the correct pedagogical and clinical context.
AI Leaders
Training health professionals since 2018
ALMA Project
Project ALMA arises as the union of the efforts of BinPar (with more than 15 years of experience in healthcare SW solutions), Editorial Médica Panamericana (with more than 70 years providing content to healthcare professionals) and AWS (a world leader in cloud services and solutions) to create a conversational assistant designed specifically for healthcare professionals .
ALMA combines the power of large language models with customization to the local medical domain , ensuring accurate, relevant and useful answers for clinical practice. Throughout this paper, we will demonstrate the relevance, excellence and superiority of ALMA over other alternatives. We will present its key performance indicators , comparisons with leading conversational models, and the features that make it a high-value strategic solution for Health Services. We will also explore how ALMA offers not only correct answers, but also didactic explanations with scientific evidence , and how its technical integration (SSO, clinical protocols, dashboards, etc.) facilitates its deployment in healthcare environments ensuring safety and regulatory compliance.
Médica Panamericana revolutionizes medical exam prep with AWS generative AIFinally, we will highlight use cases such as that of the Servei Català de la Salut (CatSalut) , illustrating the real-world application of ALMA in public healthcare, and why choosing this solution means leading innovation with ethical and sovereign AI.
Key KPIs of the ALMA Project
Quantitative results that endorse its excellence
The following is a summary of the Key Performance Indicators (KPIs) of the ALMA Project, reflecting its impact and effectiveness from the pilot to the present:
-
Accuracy in clinical responses
98% accuracy in the MIR 2025 benchmark, equivalent to 200 out of 204 questions answered correctly in the official medical residency exam. This result places ALMA above any conversational model evaluated in that setting, outperforming all reasoning and next-generation models (see comparative section).
-
Number of professional users
Over 20,000 healthcare professionals (doctors, nurses, and clinical staff) have access to ALMA in the environments where it has been deployed. The user base has grown steadily month by month, reflecting the perceived trust and usefulness of the tool in daily practice, with more than half a million messages processed.
-
Average response time
ALMA provides near-instant answers, even for complex questions, thanks to its optimized architecture and dedicated resources. This ensures that queries do not delay clinical workflows: professionals receive valid information on the spot, much faster than searching through manuals or guidelines manually.
-
Initial adoption rate
65% of professionals in pilot centers integrated ALMA into their routine work during the first 3 months. This high adoption rate reflects both the tool’s ease of use (accessible via web and mobile, with an intuitive interface) and the added value perceived by healthcare workers from having an intelligent “copilot” at their side.
-
User satisfaction
Average rating of 4.7/5 in satisfaction surveys and user feedback. Professionals highlight the reliability of the answers and the improvement in their efficiency as key strengths. 90% say they would recommend ALMA to colleagues, and 78% report having learned or updated their knowledge on a topic thanks to the explanations provided by the tool.
-
Content coverage and updates
Over 1TB of medical information. Incorporation of 50+ clinical protocols and local clinical practice guidelines into its knowledge base. ALMA is updated regularly (weekly) with new medical evidence and regulations, ensuring that answers are based on the most recent and relevant healthcare information.
These KPIs demonstrate that ALMA is not just a promising innovation, but a functional reality with measurable impact. The fact that it already has a large community of users and outstanding performance in objective metrics provides assurance that this is a mature and proven solution in real-world settings, capable of scaling across new healthcare services with minimal risks and clear benefits.
Precision leadership
Comparison with Other Models
ALMA's superior performance becomes more tangible when compared to other state-of-the-art conversational AI models. In standardized tests and medical benchmarks, ALMA has consistently outperformed its direct competitors. One of the most illustrative evaluations has been the MIR 2025 exam, given its thematic breadth and clinical demands. While ALMA achieved a 98% correct score, recognized generalist models obtained lower percentages. The following table presents a detailed comparison of ALMA against some of the more advanced conversational models:
Conversational AI Model | Benchmark hits in MIR 2025 |
---|---|
ALMA (Proprietary, specialized project) | 200 |
Gemini 2.5 Pro | 190 |
Claude Sonnet 3.7 (Anthropic) | 182 |
GPT-4o | 180 |
GPT-4.5 | 179 |
Table 1. Performance comparison of ALMA versus other conversational AI models in the MIR 2025 benchmark. ALMA stands out as the most accurate solution in answering high-level clinical questions, outperforming commercial benchmark models.
As can be seen, ALMA leads in accuracy and number of correct answers. It is particularly significant that ALMA has achieved 200 correct answers out of 204 questions, an unprecedented mark that surpasses even GPT-4 (the OpenAI model that historically hovered around ~86-90% correct on similar medical exams) and Anthropic's Claude. Even compared to newer iterations - such as GPT-4.5 or Gemini 2.5 - which incorporate improvements, ALMA maintains an advantage thanks to its specialized and fine-tuned training in the healthcare domain.
Model IA | Correct | Errors |
---|---|---|
ALMA | 200 | 4 |
Miri Pro | 195 | 9 |
Deepseek Reasoner | 191 | 13 |
O3 | 190 | 14 |
Gemini 2.5 Pro | 190 | 14 |
O1 | 189 | 15 |
Miri | 183 | 21 |
Claude Sonnet 3.7 | 182 | 22 |
O1-mini | 182 | 22 |
GPT-4o | 180 | 24 |
GPT-4.5 | 179 | 25 |
Gemini 2.0 Flash | 178 | 26 |
GPT-4 Turbo | 176 | 28 |
Grok Vision Beta | 174 | 30 |
Deepseek Chat | 173 | 31 |
Grok 2 | 173 | 31 |
Grok 2 Vision | 172 | 32 |
GPT-4o Mini | 161 | 43 |
Claude Sonnet 3.5 | 181 | 23 |
Claude Haiku 3.5 New | 150 | 54 |
AWS Nova Lite | 143 | 61 |
Gemini 1.5 Flash-8B | 143 | 61 |
AWS Nova Pro | 168 | 36 |
Gemini 1.5 Pro | 167 | 37 |
AWS Nova Micro | 137 | 67 |
Claude Haiku 3.5 | 135 | 69 |
Gemini 1.5 Flash | 130 | 74 |
GPT-3.5 Turbo | 125 | 79 |
Llama 3 8B Instruct | 88 | 116 |
Llama 3 12B Instruct | 73 | 131 |
Table 2. All results. Note: Both Miri (used by PROMIR students, Miri Pro (second in the ranking) and alma (new version launched this month) are models developed by BinPar.(arXiv.org Evaluating Large Language Models on the Spanish Medical Inte...).
Top 10 IA Models by MIR 2025 hits
- 200 -
- 198 -
- 196 -
- 194 -
- 192 -
- 190 -
- 188 -
- 186 -
- 184 -
- 182 -
- 180 -
- ALMA
- Miri-Pro
- DS-R1
- O3
- Gem2.5-Pro
- O1
- Llama4
- Sonnet3.7
- O3-mini
- GPT-4o
Another key differentiator is performance consistency . During testing, ALMA showed a very low rate of variability in its responses: across multiple iterations of the same questions, it produced nearly identical results, indicating **robustness**. Models like GPT‑4, on the other hand, may vary some responses between attempts or show slight performance shifts if the question phrasing changes. ALMA is built with quality control and cross-validation mechanisms that minimize these deviations, ensuring the response provided is reliable and consistent .
In terms of capabilities beyond the evaluated domain, it's worth noting that ALMA also remains competitive. For example, in clinical text understanding, summarizing medical reports, or drafting scientific documents , ALMA performs above its competitors in generation quality, with the added advantage of understanding the healthcare context (e.g., knowing how to prioritize clinically relevant information in a summary).
To conclude this comparative section: ALMA stands out not only in numbers but in the clinical relevance of its responses . It has demonstrated superior performance over the most advanced conversational AIs on the market, which grants it a badge of technical excellence . For a healthcare institution, this translates into the reassurance of having the best available tool in conversational AI, with objective evidence of its superiority where it truly matters: accuracy and quality in real medical contexts.
Providing the correct answer is fundamental, but in the healthcare field accuracy alone is not enough : how that answer is delivered also matters. A strong point where ALMA excels is in the didactic quality and scientific foundation of its responses.
Didactic and Evidence-Based Responses
Much more than hits: pedagogy and scientific rigor
Each time a professional makes a query, ALMA not only provides the answer , but accompanies it with a clear and concise explanation of the reasoning or relevant medical information behind it. Instead of “black box” or merely telegraphic answers, ALMA produces mini-explanations in the style of an expert or clinical tutor. This includes, for example, detailing the probable cause of a symptom, listing the steps of a diagnostic protocol, or contextualizing a therapeutic recommendation based on clinical guidelines, incorporating visual content, generating diagrams, video explanations, etc.; always in a language adapted to the professional user (technical when necessary, but focused on comprehension and practical utility ).
Most importantly, ALMA makes its sources of information transparent . Each answer is accompanied by bibliographic references, links to guidelines or studies from which the evidence is extracted. This approach responds to a growing demand in the sector: recent expert consensus has emphasized that conversational health assistants must be transparent with their sources and guarantee truthful information . ALMA fully complies with this ethical and reliable AI requirement: if a physician asks, for example, about the first-line treatment in a certain rare pathology, ALMA will cite the official clinical guideline or reference study that supports the given recommendation. This allows the practitioner to verify the information quickly if desired, increasing confidence in the tool.
The presence of contrastable references not only provides veracity but also has a didactic effect: the user can go deeper into the original source (for example, by opening the article or the page of the referenced guide) to broaden his knowledge. In generic conversational models, obtaining reliable references often requires additional queries or is not possible directly. ALMA has this functionality natively integrated, making it an evidence-based assistant .
Another aspect of the didactics is that ALMA is designed to justify its answers . If asked “Why do you recommend such a procedure?”, it does not shy away from explanation: it can detail the clinical criteria, expected outcomes and associated risks, practically teaching the questioner. This contrasts with other AIs that, if not specifically programmed, might give a short answer or even show insecurity at the request for explanation. ALMA, on the other hand, has a solid base of structured medical knowledge and a layer of clinical reasoning that allows it to argue its conclusions logically.
The combination of accuracy + explanations has an immediate benefit in care settings: it improves the quality of decisions. A practitioner who understands the “why” behind an ALMA recommendation can apply that knowledge with greater conviction and accuracy. For example, he or she will not only know which antibiotic is the antibiotic of choice for endocarditis, but why (spectrum needed, evidence in guidelines, etc.), which reinforces informed decision making.

In short, ALMA offers triple quality answers: correct, didactic and justified with evidence . This comprehensive quality distinguishes it markedly from other wizards, and turns every interaction into an opportunity not only to get an answer but to learn and reinforce knowledge . For Health Services, this feature means that the tool adds educational and continuous improvement value, ensuring that the information circulating in the organization is always supported by the best available scientific evidence.
Value Added: On-the-Job Learning
Continuing education integrated into clinical practice
ALMA's didactic capacity mentioned above translates into a strategic added value : the tool facilitates and enhances the continuous learning of professionals while working . In the busy clinical day-to-day, healthcare staff often do not have time for refresher courses or to calmly review the literature; ALMA helps to close that training gap in a natural way .
Each interaction with ALMA can be seen as a micro-training session. For example, a young family physician asks ALMA how to handle a certain complex case that comes into his office. ALMA will not only tell him the course of action to follow, but will provide him with the reasoning and references. This doctor not only solves the specific doubt to attend his patient, but also acquires knowledge applicable to future cases . Over time, after dozens of consultations with ALMA, his or her store of updated information grows. In effect, ALMA acts as a virtual mentor to the professional, reinforcing his or her skills with each response. This on-site learning has several positive implications:
-
Reduction of clinical variability
By providing answers aligned with official guidelines and consensus, ALMA helps standardize practices. All professionals—from residents to seasoned specialists—receive the same evidence-based recommendation when asking about a topic. This reduces disparities in care and ensures that the most up-to-date knowledge permeates throughout the organization. -
Constant updating
Since ALMA stays current with new evidence, professionals are indirectly kept up to date every time they consult it. Instead of waiting to read a newsletter or attend a conference, the latest recommendations can reach the consultation through ALMA. For example, if a new drug emerges or a protocol is updated, ALMA will incorporate that information, and the next time someone asks about it, the change will be shared. -
Accelerated learning curve for new professionals
For residents, newly incorporated doctors, or staff in training, ALMA is an invaluable educational support tool. It allows them to resolve doubts independently, at any time, without the fear of “bothering” a superior, and to receive reliable answers. This speeds up their learning process and builds their confidence, knowing they have a trustworthy resource at all times. -
Continuing professional development (CPD)
Many institutions require credits or evidence of ongoing education. ALMA’s use could be integrated into CPD programs, where interactions with the tool (e.g., number of relevant queries made, integrated quizzes) contribute to the professional’s training record. -
Initial user testimonials confirm this value
“I feel like I learn something new every week with ALMA,” “Now I better understand the rationale behind clinical decisions I used to make from memory,” are frequent comments. This dimension of on-the-job training is difficult to quantify in terms of immediate ROI, but in the medium and long term, it translates into a more knowledgeable, confident, and efficient healthcare workforce—something priceless for any health service.
ALMA is not just a tool that resolves doubts; it's a system that organically raises the knowledge level of the organization. By adopting ALMA, a Health Service is also investing in its human capital: the platform functions as a continuous training program, embedded into the workday without interrupting it, and aligned exactly with the real-world cases and needs that arise in practice. This represents a paradigm shift in how continuing medical education can be delivered, with a direct impact on the quality of care.
Technical Integration and Content Customization
SSO, local protocols and centralized monitoring
For a conversational AI solution to be viable in a real healthcare setting, it is not enough for it to be smart; it must be easy to integrate, secure, and adaptable to the specific needs of each institution. ALMA has been designed from the ground up with an adaptable architecture in mind, tailored to the requirements of Health Services in terms of IT, security, and customization. Below are the key technical aspects that make ALMA a solution ready for deployment in complex clinical environments:
-
Integration with Single Sign-On (SSO) and corporate systems
ALMA seamlessly integrates with the organization’s existing authentication mechanisms. It supports standard protocols like OAuth2/OpenID, SAML, LDAP/Active Directory, allowing professionals to access the platform using their corporate credentials . This simplifies adoption (no need to manage separate users) and strengthens security, ensuring only authorized personnel can access the tool. For instance, in hospitals where staff log in through an internal portal or intranet, ALMA appears as an additional option without requiring reauthentication, thanks to SSO. Integration with directories also enables permission management : defining which user profiles can access specific features.
-
Customization and ingestion of local content (clinical protocols)
One of ALMA’s biggest differentiators is its ability to absorb and utilize each Health Service’s proprietary knowledge . The system includes a document ingestion module that enables uploading local clinical protocols, internal practice guidelines, procedure manuals, regional regulations, etc. These are indexed and fine-tuned within ALMA so that when related questions arise, the assistant prioritizes answers based on "what is done here." For example, if a doctor asks “What’s the stroke protocol here?”, ALMA will cite the specific protocol from that hospital. This ensures alignment with local clinical policy and increases the relevance of answers. Each institution can have its own customized “instance” of ALMA. Maintenance is simple: when a protocol is updated, it’s reloaded and the model adjusts its answers accordingly.
-
Data privacy and security compliance
ALMA is designed to operate in environments handling sensitive information, fully complying with data protection regulations (LOPD/GDPR) . Queries sent to ALMA never leave the authorized environment , thus avoiding the risk of potentially sensitive data (e.g., identifiable clinical information) being sent to uncontrolled third-party servers. Unlike standard services like ChatGPT, where confidentiality may be a concern, ALMA operates in a closed and secure environment . No user interaction is used to retrain global models without authorization. Audit mechanisms are also implemented: every query and response can be logged (anonymized if preferred) for quality control and compliance. ALMA can also mask or filter sensitive information in answers when integrated with patient data systems, ensuring strong privacy protections.
-
Monitoring dashboards and usage metrics
For administrators and decision-makers, ALMA provides a control panel to visualize in real time and in periodic reports how the tool is being used. These dashboards show KPIs such as: number of daily queries, most asked topics (categorized by specialty or type), peak usage hours, average response time, success rate (e.g., how often users rate answers as useful), and more. This analytics helps identify training needs , measure impact (e.g., increased consultation after ALMA adoption), and demonstrate ROI (e.g., estimating hours saved by avoiding manual searches). Alerts on errors or issues are also logged, helping the tech team continuously improve the model and close the feedback loop.
From a technical and implementation standpoint, ALMA meets all key requirements to be adopted institutionally without friction: it integrates with existing IT infrastructures, respects security and privacy, adapts to each organization’s content, and gives managers the tools to monitor and maximize its use. This makes ALMA a robust and scalable solution ready for deployment in large regional health services, nationally, or across hospital networks.
ALMA
As a strategic commitment for Health Services
Healthcare systems are at a turning point where digital transformation and artificial intelligence will be crucial to tackle current and future challenges (population aging, chronic conditions, need for efficiency, constant updates, etc.). In this context, ALMA Project is presented not only as a technological tool but as a comprehensive strategic commitment:
-
Proven excellence
ALMA has shown outstanding performance in accuracy and reliability, surpassing leading generalist AI models. Its 200/204 correct answers in the MIR 2025 benchmark are clear proof of its level of technical excellence , placing it at the forefront of conversational AI in medicine. This accuracy translates into better informed clinical decisions , reduced risk of error, and ultimately, higher quality patient care. With ALMA, healthcare managers can be confident they are equipping their teams with a top-tier assistant .
-
Human talent enhancer
Far from replacing professionals, ALMA empowers and trains them . It offers educational and evidence-based responses that enrich users’ knowledge with every interaction. Thus, investing in ALMA is investing in the continuous training of healthcare staff in a practical, job-oriented way. Better-supported and trained staff leads to improved care practices and higher job satisfaction. In an environment where talent retention is critical, tools that facilitate work and learning can make a real difference.
-
Seamless integration and security
ALMA complies with interoperability, security, and privacy requirements demanded by public institutions. Its easy integration with SSO and existing systems minimizes implementation efforts. The ability to integrate proprietary protocols and control deployment ensures that the organization's internal knowledge remains protected, preserving the institution’s technological sovereignty . At a time when administrations aim to reduce external dependencies and ensure confidentiality (ethical AI), ALMA offers a solution aligned with these principles (as highlighted by expert consensus calling for transparency and accuracy in sources). This greatly facilitates justification and acceptance of ALMA in ethics committees, patient safety units, and IT audits.
-
Ready for public procurement and scalability
From a technical-commercial perspective, ALMA is ready to be tendered and deployed at scale. Its unique features can be articulated as high-value awarding criteria (e.g., validated accuracy level, automatic bibliographic referencing, local content training capabilities, etc.), giving it an edge in calls for the best possible solution. Additionally, its licensing model and architecture allow scaling from pilot projects to full regional health service deployments with sustainable costs and dedicated support. Investment in ALMA is justified not only by its capabilities but by its expected return: time savings for professionals, fewer second-opinion consultations, standardized clinical criteria, etc. Each of these impacts has significant economic and healthcare value that can be substantiated in any project brief or tender document.
ALMA represents a strategic solution for the future , that is already a present reality. Its relevance lies in combining the best of artificial intelligence (speed, massive access to knowledge, automation) with the best of human judgment (clinical context, empathy, and pedagogy), serving as a bridge between both. Choosing ALMA means leading by example in the responsible adoption of AI in healthcare. It means equipping professionals with a cutting-edge tool that improves their work and, therefore, the care provided to citizens. And it means aligning with a vision of an innovative, efficient, and secure healthcare system.
ALMA is not just a solution — it's a technology partner in the mission to elevate the quality of care. Implementing it means taking a firm step towards meaningful digital health transformation , one that respects science, people, and privacy. For all these reasons, ALMA stands out as a winning strategic commitment: an investment with multiple returns (clinical, educational, and organizational) and an initiative that places the health service at the forefront of innovation.
In a world where AI is redefining industries, we ensure that healthcare stays one step ahead, using artificial intelligence to care better, learn every day, and shape a future where technology and humanity walk hand in hand for the benefit of public health.