Grounded knowledge research

Keeping complex knowledge current, traceable, and bounded.

Aprilio is a multidisciplinary research effort developing structure-based retrieval for knowledge that changes faster than conventional AI systems can reliably follow.

Collaborate on research Explore the method

Working prototype

Internal evaluation

External validation invited

Aprilio research system Working prototype

Guideline

source structure

Live update

new evidence

Clinical study

emerging signal

Ontology

Source map

structure + meaning

Factumbounded

Current · linked · supported

Animated research workflow · not a clinical decision tool

Research agenda

Three questions guide the work.

Our work asks how AI can navigate changing medical knowledge without losing source structure, recency, or the boundary of available evidence.

01Open question

How can AI preserve the structure of a medical source?

We study retrieval systems that navigate sections, pathways, and relationships instead of flattening knowledge into statistically similar fragments.

Source structure

02Open question

How should a system respond when the evidence changes?

We examine how clinical guidance, new approvals, and emerging evidence can be reconciled without silently rewriting established consensus.

Change detection

03Open question

Can an answer expose the boundary of what is known?

We design outputs that remain tied to their evidence and stop when the available sources cannot support a claim.

Evidence bounds

Research prototype

Structure before generation.

Grounded Adaptive Retrieval builds a source-specific map, resolves a question against that structure, and produces a Factum only when the retrieved evidence can support it.

Map

Represent the source hierarchy, entities, and decision logic.

Reconcile

Resolve clinical intent and check the source set for change.

Bound

Return source-linked claims or expose where evidence ends.

Aprilio / Live Factum Trace 01

processing

?Clinical query

What changed in first-line AML treatment for older adults?

01Map sources

02Retrieve evidence

03Reconcile updates

04Issue Factum

01 / Source field

2 live inputs

GUI

GUIDELINE

ESMO · Acute Myeloid Leukemia

v3.2026 · 214 pages

Waiting for trace

LIV

LIVE UPDATE

FDA approval notice

13 May 2026 · newer than source

Waiting for trace

Resolved intent

AML1st lineolder adultnon-intensive

02–03 / Ontoharness

Map sources

Cartographer

Mapped 14 sections to the AML pathway

Semantic annotator

Resolved intent: 1L · older adult · non-intensive

Architect

Separated guideline consensus from new evidence

04 / Bounded answer

Assembling

Building a Factum

Claims appear only after structure, recency, and provenance checks complete.

pages mapped

sources reconciled

unsupported claims

Inspect the full architecture

Evaluation

Three questions. One changing evidence problem.

We compared five frontier models with Aprilio's research prototype across three oncology questions where recency, context, and evidence boundaries materially change the answer.

HCC evaluation

“What are preferred first-line systemic treatment options for hepatocellular carcinoma?”

What foundation models returned

Perplexity

Recommends camrelizumab + rivoceranib as a first-line option

FDA rejected twice (May 2024, March 2025)

DeepSeek

Includes sintilimab in preferred options

Unavailable in the United States

What Aprilio retrieved

For unresectable or metastatic HCC with preserved liver function, retrieves atezolizumab + bevacizumab, durvalumab + tremelimumab, and nivolumab + ipilimumab as preferred first-line options. Lenvatinib and sorafenib remain alternatives when immunotherapy or VEGF inhibition is unsuitable.

Preferred regimens

Atezolizumab + bevacizumab, durvalumab + tremelimumab, and nivolumab + ipilimumab.

Outcome evidence

Surfaces OS and response evidence from IMbrave150, HIMALAYA, and CheckMate 9DW.

Clinical boundary

Retains lenvatinib or sorafenib when immunotherapy or VEGF inhibition is unsuitable.

Source basis

ASCO guideline + NCI PDQ + FDA approval, updated April 2025

Current approval surfaced: nivolumab + ipilimumab, April 11, 2025

Open the complete captured response, including tables, evidence notes, and references.

The prototype separates preferred immunotherapy combinations from established alternatives and surfaces the approval date that changed the treatment landscape.

Comparison set

5 frontier models

Clinical scope

3 oncology questions

Validation status

External review invited

Methods and research notes

Open the full evaluation

Research note

The Neurosymbolic Advantage: Why Structured Data Beats Bigger Models

Failure study

What Happens When AI Recommends an FDA-Rejected Drug?

Methods essay

Why We Built Grounded Adaptive Retrieval

Research collaboration

Help us study retrieval where the knowledge is real.

We are looking for clinical, academic, and knowledge-source partners who want to evaluate grounded retrieval within a clearly governed research setting.

Discuss a collaboration Meet the team

Knowledge sources

Study a governed source

Evaluate how structure-aware retrieval behaves across a trusted medical knowledge base.

Clinical research

Co-design an evaluation

Define clinically meaningful questions, failure modes, and review criteria with domain experts.

New domains

Validate another specialty

Test whether the approach transfers beyond our initial work in oncology.

Knowledge change

Study what changed

Measure how the system responds when guidance, evidence, or source organization evolves.

Governance first

Every collaboration should define permitted use, access, retention, attribution, publication, and data handling before research begins.