Platform

The platform behind every analysis.

A biomedical knowledge graph with more than 10 billion nodes and 25 billion edges, specialized AI agents, and scientists working over the same evidence.

Updated October 20, 2026

Talk to our team See plans

Nodes in the knowledge graph: 10B+
Edges connecting them: 25B+
Terminated trials in our published analysis: 3,180

Knowledge graph scale as of September 2026.

From sources to delivery

How the platform produces an analysis.

Sources

Biomedical evidence

Trials, literature, patents, drug labels, conferences, and company filings.
Connected evidence

Epistemic Knowledge Graph

10B+ nodes · 25B+ edges

As of September 2026.

Entities and relationships with their source records.
Agents

Specialized AI agents

Search, extract, link, and compare the evidence, then draft for review.
Delivery

Finished analysis

Reports, decks, dashboards, or data, refreshed on an agreed cadence.

Scientists work throughout the investigation

They scope the question, inspect sources, challenge assumptions, interpret the findings, and own the conclusions you receive.

Epistemic Knowledge Graph

10 billion nodes of drugs, targets, trials, and their evidence.

Knowledge graph scale as of September 2026.

The graph connects drugs, targets, genes, diseases, clinical trials, companies, publications, patents, regulatory labels, and conference abstracts. Mentions in each source are linked to canonical entities, including synonyms and alternate names, so a question about one asset reaches the trials, papers, and filings that refer to it.

Relationships keep a link to the source records they came from. That is what lets scientists and your reviewers trace a claim back to its evidence.

Sources the graph draws on

ClinicalTrials.gov and other trial registries
PubMed, biomedical journals, and conference abstracts
FDA, EMA, and other regulatory documents and drug labels
Global patent databases
Company filings, press releases, and market data

The sources for each analysis are selected for its scope.

Connected entities, relationships, and source records form a reusable evidence foundation.

More than a general-purpose AI response

General-purpose models are useful for drafting and summarizing. Strategic intelligence also requires a structured evidence base, source traceability, maintained analysis, and accountable scientific review.

Epistemic AI

Evidence organized as reusable entities, relationships, and source records
Analysis grounded in the scoped Epistemic Knowledge Graph
Provenance carried into analysis and review materials
Finished competitive landscapes, TPPs, conference intelligence, and related analysis
Life-science experts inspect evidence and own conclusions

General-purpose workflow

Built around a prompt and response
Evidence depends on supplied context
Structure is recreated for each task
The user evaluates the response

Agent architecture

Specialized agents, each built for one kind of work.

Instead of one general-purpose model answering from memory, the platform runs agents with defined jobs over indexed sources and the knowledge graph. Scientists direct the work and review what the agents produce.

Source agents

Search indexed clinical trial records, scientific literature, conference abstracts, press releases, and drug labels, and return the passages and records behind each finding.

Query agents

Turn a question into structured queries over trial, label, and drug databases, so counts and comparisons come from the data rather than from a model's recall.

Extraction and linking

Pull indications, endpoints, dosing, and outcomes from documents and link each one to the graph's canonical entities.

Research agents

Plan multi-step investigations across the graph, indexed sources, and the web, and reconcile what they find.

Synthesis agents

Assemble target product profiles and competitive landscapes from the gathered evidence, with sources attached for review.

Evaluation

We test agents against curated evaluation sets and trace their runs, so we can measure how each change to the system affects results.

Evidence and provenance

Trace the record behind the analysis.

Provenance stays with the evidence as it moves through the agents and scientific review. Depending on the scope and format, review materials include source links, evidence registers, assumptions, and identified gaps.

Scientists are part of the investigation, not a final approval step. They scope the question, inspect the sources, challenge the reasoning, and own the conclusions your team receives.

Published research

A decade of trial terminations, in Nature Reviews Drug Discovery.

Using EpistemicGPT, our biomedical research assistant, and the Epistemic Knowledge Graph, we analyzed 3,180 terminated Phase 2 and Phase 3 trials from 2013 to 2023. The analysis found that strategic and business decisions have overtaken scientific causes as the leading driver of late-stage terminations.

Analysis of phase II and phase III clinical trial terminations from 2013 to 2023, Nature Reviews Drug Discovery, December 18, 2025. doi:10.1038/d41573-025-00208-6

EpistemicGPT connects a generative model to the curated knowledge graph. We introduced it in 2023.

Read our summary of the findings

Compounding

Each question starts from the last one.

Evidence gathered for one analysis is already connected in the graph. When the next question overlaps, whether it's a new indication, competitor, or conference, the work starts from that evidence instead of from a blank search. Scientists confirm what is still current, what needs new research, and what the agreed terms permit.

Our AI agents gather new evidence and update the analysis, with scientists reviewing the changes on an agreed cadence. Next, we are building a workspace where client teams can explore and extend the connected evidence themselves.

Your next decision starts here

What decision is
your team facing?

Bring us your question. We reply within one business day and help define the analysis that moves you forward.

Talk to our team