Artificial intelligence is revolutionizing medical coding, an essential process for billing, research, and clinical management. However, traditional language models require large volumes of annotated data that are expensive and difficult to obtain. This is where OntoBook marks a turning point: by converting medical ontologies — hierarchical and causal structures of codes like ICD-10, CCAM, or ATC — into synthetic textbooks, it efficiently trains language models without relying on real clinical data. This methodology not only improves performance on coding tasks but also opens the door to customized solutions for hospitals, insurers, and healthcare providers.
OntoBook's process is divided into three phases. First, random walks are performed on ontology graphs, capturing hierarchical and causal relationships between codes. Second, a large language model reformulates those walks into fluent prose, mimicking the style of a textbook. Finally, that synthetic text is used to train an encoder (149M-parameter ModernCamemBERT) with two simultaneous objectives: masked language modeling (MLM) and relation prediction between code pairs. Results on French benchmarks (FRACCO, Cantemist-FR, Distemist-FR) are compelling: +2.5 micro-F1 on FRACCO and +8.0 micro-F1 on Distemist, demonstrating that alignment between objectives is key — when misaligned, performance drops by 30 points.
For a company like Q2BSTUDIO, specialized in custom software development and artificial intelligence, OntoBook represents a perfect use case. Instead of relying on restricted clinical data, synthetic corpora can be generated from existing ontologies, training models that are then integrated into cloud AWS/Azure systems for secure scaling. Além disso, OntoBook's architecture allows for AI agents that automate code review, reducing errors and administrative costs. The combination of ontologies, LLMs, and multi-task learning is exactly the kind of innovation healthcare organizations need to modernize their workflows.
From a business perspective, medical coding is a bottleneck: it requires specialized personnel, is error-prone, and consumes valuable time. OntoBook, by offering high-quality synthetic textbooks, enables training models that can act as intelligent assistants. At Q2BSTUDIO we have seen how similar solutions, based on synthetic data generation and multi-task learning, can be applied to other domains — from automated billing to electronic health record management. Our focus on custom applications allows us to adapt these techniques to each client, integrating proprietary ontologies and language models in cloud environments with high cybersecurity standards.
One of the most relevant findings from OntoBook is the importance of aligning training objectives. When the model learns to predict code relationships using different data than the masked language modeling task, performance plummets. This reinforces the need for coherent data pipelines, something Q2BSTUDIO applies rigorously in our AI projects. We work with platforms like Power BI to visualize data quality and with AI agents that monitor model stability in production. Cybersecurity also plays a critical role: synthetic data avoids exposing sensitive patient information, and our cloud architectures include encryption and granular access control.
The future of medical coding lies in intelligent automation, and OntoBook is a solid step in that direction. By releasing 1.3 million LLM-reformulated textbooks, researchers can explore new frontiers in medical representation learning. At Q2BSTUDIO we believe that the convergence of ontologies, artificial intelligence, and cloud computing will transform how healthcare information is managed. Whether implementing an automatic coding system for a hospital or developing an AI-agent-based diagnostic assistant, our experience in custom software solutions allows us to accompany organizations at every stage of the process.
In summary, OntoBook demonstrates that synthetic data generated from ontologies can be as effective as real data for training language models in specialized domains. This approach not only reduces costs and accelerates development but also improves accuracy and security. For healthcare companies seeking innovation, combining tools like OntoBook with Q2BSTUDIO's expertise in AI, cloud, and business intelligence is a winning bet. We invite industry professionals to explore how these technologies can be applied to their specific cases, contacting our team to discuss customized solutions.
