Eldrix’s Hermes project is a critical piece of infrastructure for AI agents operating in the healthcare domain. For an agent to be reliable in a clinical setting, it must ground its reasoning in established medical ontologies like SNOMED CT to avoid hallucinations of medical codes or nonsensical relationship mappings. Hermes provides the deterministic source of truth that agents query via a Graph-API, allowing them to validate clinical concepts and explore hierarchical relationships between diseases, symptoms, and treatments.
In the context of the agent stack, Hermes operates as a specialized knowledge retrieval layer. It is particularly valuable for Retrieval-Augmented Generation (RAG), where an agent can use the Graph-API to expand a user's query into its constituent clinical parts or verify that a suggested diagnosis adheres to medical logic. By providing a high-performance, open-source method to navigate medical knowledge graphs, Eldrix enables the creation of domain-specific agents that are both semantically aware and clinically accurate.
Eldrix, led by clinician and developer Mark Wardle, focuses on the structural challenges of medical data through its primary project, Hermes. The platform is built specifically to address the complexity of SNOMED CT, the world’s most comprehensive clinical terminology system. SNOMED CT is distributed in the RF2 format—a collection of relational files that are notoriously difficult to query directly without significant pre-processing. Hermes acts as a bridge, transforming this raw data into a high-performance terminology service that supports full-text search and complex hierarchical navigation.
At its core, Hermes uses a stack optimized for both search and relational depth. It employs Lucene for indexing medical descriptions and LMDB (Lightning Memory-Mapped Database) as a high-performance key-value store for clinical concept data. This architecture allows the system to provide near-instantaneous lookups and traversals across millions of medical relationships. The implementation in Clojure is a deliberate choice, reflecting the language's suitability for handling the immutable, data-centric nature of clinical ontologies.
A distinguishing feature of the project is the com.eldrix.hermes.graph namespace. This Graph-API uses discrete resolvers via Pathom, a library for building data-driven APIs. In the context of SNOMED CT, this is a significant departure from standard keyword-based search. It allows software to treat clinical terminology as a navigable graph where concepts are nodes and relationships—such as "is a," "finding site," or "causative agent"—are edges. This capability is vital for applications that need to validate clinical reasoning or perform semantic expansion on medical records.
Beyond basic lookups, Hermes implements the SNOMED Expression Constraint Language (ECL). ECL is a formal language used to define sets of clinical concepts based on their properties. By supporting ECL, Hermes enables developers to write powerful queries, such as identifying all concepts related to "viral pneumonia" that are also "infectious diseases of the lung." This level of precision is necessary for clinical decision support systems and advanced medical research tools.
Based in the United Kingdom, Eldrix operates as a specialized consultancy, with the Hermes project maintained as an open-source library. This approach allows the healthcare community to scrutinize the implementation and contribute to its development, which is critical in a field where data accuracy is paramount. The project includes built-in importers for SNOMED distributions, a Verhoeff check digit algorithm for identifier validation, and tools for handling language and locale matching.
While many terminology servers are marketed as large-scale enterprise suites, Eldrix focuses on the developer experience, providing a library jar that can be embedded directly into JVM-based applications. This allows for local, high-speed access to clinical data without the latency or privacy concerns associated with calling external cloud-based terminology APIs. For organizations building next-generation healthcare platforms, Hermes provides the deterministic backbone required to make sense of the vast, interlocking world of medical concepts.
A terminology service and Graph-API for SNOMED CT.
Eldrix is hiring