A clinical Q&A agent that answers questions across conditions, interventions, and evidence — built for researchers, clinicians, and health-curious developers.
axiom-md is a LangChain-powered CLI agent that answers structured clinical questions across a wide range of disease areas and interventions. It is designed to mirror how a knowledgeable clinician reasons through evidence: starting with a question, identifying relevant terminology, and synthesizing answers from authoritative sources.
Ask it about a drug-condition interaction, a diagnostic criterion, a common patient question, or an evidence gap — it routes the query, retrieves relevant context, and returns a structured, sourced answer.
- Multi-domain clinical Q&A — handles questions across conditions (common to rare), interventions, diagnostics, and patient-facing topics
- Structured output — answers include condition name, relevant clinical ontology codes (ICD-10, RxNORM, LOINC where applicable), evidence summary, and source citations
- PubMed integration — retrieves abstracts from PubMed for evidence-backed responses
- Ontology-aware — maps clinical concepts to standard terminologies to support downstream search and retrieval
- Extensible pipeline — designed to be swapped into larger evidence generation workflows
$ python agent.py
axiom-md > What are the treatment options for Fuchs' Corneal Dystrophy,
and at what stage is surgery typically indicated?
Agent: Querying clinical knowledge base...
→ Found entry: Fuchs' Endothelial Corneal Dystrophy
Agent: Searching PubMed for supporting evidence...
→ Query: "Fuchs endothelial corneal dystrophy treatment DMEK DSAEK"
→ Retrieved 3 abstracts (PMIDs: 33248268, 25892019, 32511983)
Agent: Mapping clinical ontology codes...
→ ICD-10: H18.59 (Other hereditary corneal dystrophies)
→ RxNORM: 8262 (sodium chloride 5% ophthalmic solution)
→ LOINC: 71486-3 (corneal endothelial cell density, specular microscopy)
────────────────────────────────────────────────────────────────────────
Condition: Fuchs' Endothelial Corneal Dystrophy
ICD-10: H18.59
RxNORM: 8262 (sodium chloride 5% ophthalmic)
LOINC: 71486-3 (specular microscopy — endothelial cell density)
Summary:
Fuchs' Endothelial Corneal Dystrophy is a bilateral, progressive
degeneration of the corneal endothelium characterized by guttae
formation and eventual endothelial cell failure. Management is
stratified by disease severity.
Early-stage disease (guttae without edema): Managed conservatively
with hypertonic saline drops (NaCl 5%) to reduce epithelial edema.
Patients typically report morning blur that clears through the day
as the epithelium dehydrates — this pattern is diagnostically useful.
Surgical indication: Endothelial keratoplasty is indicated when
corneal edema produces visually significant impairment unresponsive
to medical management. DMEK (Descemet Membrane Endothelial
Keratoplasty) is the current standard of care, demonstrating faster
visual recovery and lower rejection rates vs. DSAEK in controlled
trials. DSAEK remains appropriate where DMEK tissue or surgical
expertise is unavailable. Descemet stripping only (DSO/DWEK), without
donor graft, is an emerging option for select cases with sufficient
peripheral endothelial reserve.
Sources:
- Ong Tone et al. (2021). Fuchs endothelial corneal dystrophy:
The vicious cycle of Fuchs pathophysiology.
Progress in Retinal and Eye Research. PMID: 33248268
- Price et al. (2015). Descemet membrane endothelial keratoplasty:
Prospective multicenter trial of visually significant guttae.
Ophthalmology. PMID: 25892019
- AAO Preferred Practice Pattern: Corneal Ectasia (2023)
────────────────────────────────────────────────────────────────────────
axiom-md >
User query (CLI)
│
▼
LangChain agent
│
├──► Clinical knowledge base (local structured Q&A library)
│
├──► PubMed API (abstract retrieval via Entrez)
│
└──► Ontology mapper (ICD-10 / RxNORM / LOINC lookup)
│
▼
Structured response
(condition, codes, summary, sources)
| Layer | Technology |
|---|---|
| Agent framework | LangChain |
| LLM | OpenAI GPT-4o |
| Literature retrieval | PubMed Entrez API (Biopython) |
| Ontology mapping | Custom ICD-10 / RxNORM lookup layer |
| Interface | Python CLI |
Prerequisites
- Python 3.10+
- OpenAI API key
Install
git clone https://github.com/yourusername/axiom-md.git
cd axiom-md
pip install -r requirements.txtConfigure
cp .env.example .env
# Add your OPENAI_API_KEY to .envRun
python agent.pyActive development. Core Q&A pipeline and PubMed retrieval are functional. Ontology mapping layer and structured output schema are in progress.
Planned:
- Expanded clinical Q&A library across 50+ conditions
- LOINC code mapping for lab and diagnostic queries
- Metatag output for search and retrieval integration
- ClinicalTrials.gov integration for active trial retrieval
- Export to structured JSON / markdown for downstream pipeline use
This project began as an exploration of LangChain agents for multi-domain Q&A. As the architecture matured, the focus shifted toward clinical evidence — a domain where the need for structured, sourced, rapidly retrievable answers is acute. axiom-md is designed as a building block for larger evidence generation pipelines, not a standalone clinical decision support tool.
This project does not provide medical advice and is not intended for clinical use.
auris— HIPAA-aware voice pipeline for clinical call transcription and structured outcome extractionpriorx— RAG-powered clinical evidence prioritization engine
MIT