Every fact, connected to the page, event, provider, and record it came from. BETA
The Medical Record Knowledge Graph converts raw pages into a structured evidence graph: documents, mentions, medical entities, events, and the relationships between them, each node holding its own provenance. Extracted facts are kept visibly separate from inferred relationships, so a reviewer always knows which edge is cited and which is a modeled connection.
From page to structured node, automatically.
Every page is segmented into documents and mentions, then normalized into medical entities and events — diagnoses, treatments, providers, dates, imaging findings — each one linked back to its exact source page. The graph is the structured layer sitting underneath the record you already read.
Relationships, drawn and labeled — extracted vs. inferred.
A fact pulled straight from a page is never presented the same way as a connection the platform has drawn between two facts. Every edge in the graph is labeled extracted or inferred, carries a confidence signal, and can be overridden by a reviewer.
One evidence layer, many outputs.
Chronology, Q&A, and summary generation already read from this same graph. This page is a distinct, more specific evidence-graph layer built on top of Medical Data API, not a replacement for it — if you want to pull the graph itself into your own tools, that's Evidence Graph API.
Cited facts and modeled connections never look the same.
No edge in the graph appears as verified unless it's genuinely cited to a source page. Extracted facts and inferred relationships are always shown as separate categories, each carrying its own confidence and override history.
And the per-case graph is never conflated with cross-case aggregate patterns — what this page builds stays scoped to the file it came from, not blended into a portfolio-wide model without a separate, explicit feature saying so.
From pages to a graph you can query and trust.
Three steps, running underneath the features you already use.
Pages are segmented into documents, mentions, and medical entities, each cited to its source.
Edges are drawn between nodes and labeled extracted or inferred, each with a confidence score.
Low-confidence and inferred edges route for human confirmation before they're relied on.
Who reads the record as a graph.
Any team that already relies on chronology, Q&A, or summaries is already reading this graph — some want to see the structure directly.
Trace a causation theory node by node, back to the page.
For law firmsSee extracted findings and inferred links kept clearly apart.
For IME orgsOne structured layer feeding every downstream file review.
For TPAsA structured evidence base underneath claims review and triage.
For carriersMedical Record Knowledge Graph, answered.
It's a structured evidence graph built from your case file: documents, mentions, medical entities, events, and the relationships between them, each node and edge holding its own citation back to the source page. It's the structured data layer underneath the chronology, Q&A, and summary features.
Medical Data API exposes the platform's raw structured record data for integration into your own systems. Medical Record Knowledge Graph is a more specific, distinct layer built on top of that data: the evidence graph itself, with entities, events, and relationships explicitly modeled and separated from the underlying extracted facts.
An extracted fact is a node taken directly from a source page — a diagnosis, a date, a provider name — cited to that page. An inferred relationship is an edge the platform draws between nodes, such as "this treatment relates to this event." The graph always displays which is which; no inferred edge is shown as if it were a cited fact.
Yes, in beta. The Medical Record Knowledge Graph is live and testable now on your own case files; we're refining it hands-on with early customers, and if your use case is a good fit we'll work with you directly.
Inside the platform, yes — the graph powers chronology, Q&A, and summary generation today. A dedicated external API surface for pulling the graph into your own tools is Evidence Graph API, a related capability also live in beta.
Related capabilities.
Pull this same graph into your own legal or claims tools.
A cross-case fingerprint built on top of this same evidence layer.
The broader structured record API this graph layer builds on.
One of the live features this evidence graph already powers.
See your own record as a structured, cited graph.
Join the beta and we'll build the graph from one of your own files. Handled under our BAA; never used to train a model.