Why this exists
Healthcare questions bring together patient context, measurements, questionnaires and domain knowledge. This public MVP explores how that context can be assembled and inspected before a language model reasons over it. It is a reference implementation for engineering patterns, not a clinical product or a source of medical advice.
Architecture
The public repository separates data loading, context construction, retrieval, local model reasoning, validation and response generation. Retrieval uses a transparent NLP and rule-based scoring path; reasoning is designed to run with a local model; the safety layer applies explicit checks before the final response stage.
Key capabilities
Patient profile, bloodwork, questionnaire, question and retrieved knowledge are assembled as explicit inputs.
Tokenization, stop-word handling, intent and synonym expansion make the retrieval path inspectable.
A local model reasoning stage is kept separate from context and retrieval boundaries.
Evidence, grounding, clinical language, prescription and disclaimer checks run as explicit validation steps.
Execution exposes query generation, evidence, reasoning mode, prompt size and validation results.
Single-responsibility components make the MVP easier to test, explain and extend.
Current scope and maturity
This is an ARMIE AI Labs public reference implementation / MVP using sample data and a mock knowledge base. It demonstrates context engineering, grounded reasoning, explainable retrieval and safety validation in a modular pipeline.
The project does not claim to be a launched medical product, diagnostic system, medical advice system, compliance platform, FHIR implementation, clinically validated system or production-ready healthcare deployment. Public evidence is limited to the repository's documented code, sample fixtures and evaluation scenarios.