Healthy Aging Knowledge Lab

Healthy Aging Knowledge — About

An open, evidence-traceable knowledge graph for healthy aging · home · browse claims

Healthy Aging Knowledge is an open, evidence-traceable knowledge graph for healthy aging — a growing map of how interventions, exposures, and measures relate to outcomes in older adults, where every claim is tied back to a real, verifiable source.

concepts
claims
evidence citations

Figures are read live from the graph. Explore it on the home page.

Why it exists

Healthy-aging evidence is scattered across thousands of trials, cohorts, and guidelines. Summaries drift from their sources, and it is often hard to tell what was actually measured, in whom, and how strong the evidence is. Healthy Aging Knowledge structures that evidence as a graph you can trace: pick any relationship and follow it straight to the paper — with the sentence that supports it.

How it is built

The graph is expanded and checked continuously by two cooperating AI agents, with a human in the loop for anything uncertain.

  1. Curator agent proposes new concepts and relationships within the healthy-aging scope. Crucially, it describes its source paper (title, authors, year) but is never trusted to supply an identifier — language models hallucinate PMIDs and DOIs.
  2. Citation resolution finds the real paper by searching PubMed and Crossref and confirming the returned title matches the described one. No confident match → the claim is dropped rather than attached to a fabricated ID. Errata and peer-review artifacts are filtered out.
  3. Reviewer agent fetches the paper's actual abstract and judges whether it genuinely supports the specific claim — the right outcome, in the right population, without overstatement. It passes two gates: the source exists, and it is grounded.
  4. Human review handles the exceptions. Anything the reviewer flags (needs_refinement) waits in a console where a curator can approve, re-verify & fix, or reject it.

The pipeline runs autonomously on a schedule and is self-healing — transient lookup failures are deferred and retried rather than mistaken for missing evidence.

What's in the graph

Nodes are concepts — diseases, outcomes, interventions, exercise, nutrition, drugs, measurement scales, populations, mechanisms, and the papers themselves. Claims are typed relationships between them, carrying a direction, a population, and a certainty. Evidence attaches each claim to a citation, usually with a supporting quote.

Node colour on the graph encodes node type; edges that carry conflicting evidence are drawn in red so disagreements are visible rather than hidden.

Explore the graph Browse claims & citations

Open & built in the open

The service is deliberately small and dependency-light: a zero-framework TypeScript backend on Node, a Postgres + pgvector database for the graph and semantic search, and Anthropic's Claude models driving the Curator and Reviewer. It runs live on Cloud Run.

TypeScript · Node Postgres + pgvector Anthropic Claude Cloud Run PubMed · Crossref

Contact

Questions, corrections to a claim, or interested in collaborating? Send a message below — or find the maintainer on GitHub.

Messages are stored securely on the server; there is no public inbox. Include an email if you'd like a reply.

Not medical advice. Healthy Aging Knowledge is a research and educational tool. Claims are machine-proposed and machine- or human-reviewed, and may be incomplete, out of date, or wrong. Always consult qualified healthcare professionals for decisions about your health.