NECTAR Platform

Nutritional Evidence and Clinical Targeting of Appropriate Recipes. A knowledge-graph and retrieval-augmented platform that builds a nutrition database from recipes and food-composition data, and recommends dishes appropriate to a patient's conditions, medications, and goals.

Educational and research use only. Not medical nutrition therapy. See LICENSE-NOTES.md and nectar/docs/SDD.md Section 9 for the intended-use boundary and validation gate.

Two programs, one contract

• contract/ — the shared Neo4j hybrid (graph + vector) schema and write-back interface. The single source of truth both programs depend on. Start here: contract/DATA_CONTRACT.md.
• nutriscrape/ — the batch data engine. Builds the database. No interactive user.
• nectar/ — the clinician-facing application. Reads the database, composes per patient.
• deploy/ — rootless Podman deployment (Quadlet units, Containerfiles, compose, Caddy, RBAC).

NutriScrape writes the graph. NECTAR reads it and performs exactly one gated write-back. The dependency direction is fixed and enforced by Neo4j roles (deploy/neo4j/init-roles.cypher).

Working in Claude Code

Open the repo root so the root CLAUDE.md loads. It imports the contract and points at each program's CLAUDE.md. The invariants in those files are non-negotiable; they encode the clinical safety design. Build order and acceptance criteria are in each program's docs/PDD.md Section "Phased Task Plan".

Quick start (development)

cp .env.example .env
cd deploy/compose && podman-compose up -d neo4j ollama       # infra
# build + run each program per its Makefile; see nutriscrape/ and nectar/

Production deployment: deploy/README.md.

Authorship and license

Author: Ronald Rodriguez, MD, PhD · UT Health San Antonio.

Copyright 2026 UT Health San Antonio. Licensed under the Apache License, Version 2.0 (see LICENSE and NOTICE). The code license covers code only; third-party data-source terms are in LICENSE-NOTES.md.