Technology

The technology behind AllSci

Proprietary scientific AI that atomizes the world's science, links it to canonical entities, and grounds every answer in the evidence.

The stack

Technologies that support real science

AllSci has developed a set of big data, ontology management, and LLM technologies that integrate scientific knowledge, add new knowledge, and make it usable.

01

Entity resolution

Every person, org, and concept resolved to one identifier — the graph everything is built on.

02

Atomization

Proprietary AI generates hypotheses, questions, and results from content.

03

Evidence engine

Every hypothesis is tied to supporting and refuting evidence across the corpus.

04

Semantic graphs

Atoms, entities and concepts in semantic graphs — traverse science in logical steps.

05

Chat

Ask AERIS in plain language and get answers grounded in the graph.

06

Agents

AI that guides researchers to results and deliverables without usurping autonomy.

07

AI peer review

Agents for reviewing scientific work — instant first-pass checks alongside the community.

08

News generation

Data-driven science news leveraging agents, workflows, and human editors.

Entity resolution

Entity resolution at scale

AllSci developed a series of technologies to map every organization, person, and concept to a canonical ID — powering a universal entity graph that enables accurate search, profiles, analytics, and AI.

ResearcherOrganizationDrug
Atomization

We turn scientific content into logical atoms

AllSci developed a technology to generate hypotheses, questions and findings from scientific text opening up a new, better, and more natural way to engage with knowledge.

Mechanisms of anti-CD20 therapy in relapsing MS
Nakamura et al. · Nat. Neurology · 2024
Hypothesis
Question
Result
Full-text paper
→
HypothesisB-cell depletion lowers relapse rate ✦ grounded
QuestionDoes anti-CD20 affect disability progression? ✦ p.4
Result46% fewer relapses vs placebo ✦ Table 2
Grounded atoms
→
anti-CD20RRMSHypothesisQuestionResult
Connected graph
Each atom links back to the exact source span · confidence surfaced as the “AERIS Drafted” label
Evidence engine

Every hypothesis, weighed against the evidence

AllSci developed a technology to gather supporting and refuting evidence for any hypothesis across the scientific corpus with full citations and linking.

HypothesisB-cell depletion lowers relapse rate in RRMS
18 supporting4 refutingfrom 22 sources
SupportsNEJM 2022 · 46% fewer relapses vs. placebo
SupportsLancet 2023 · effect sustained at 24 months
RefutesCohort 2021 · no benefit in progressive MS
Evidence grounding

~4× better at grounding claims in evidence.

General models assert; AERIS cites. On the Evidence-Grounding axis of ASC-QA-Val, AERIS answers stay tied to real, retrievable sources — including the refuting ones.

frontier best · 32.1
17.8%
GPT-5.2
32.1%
Opus 4.7
69.2%
AERIS-discovery
Semantic graphs

See how everything connects

AllSci developed semantic visualizations to enable researchers to view scientific atoms, entities and concepts in new and revealing ways.

HypothesisB-cell depletion lowers relapse rate in RRMS
12 related questions34 results7 drugs & targets5 fields
Chat

Ask AERIS, grounded in the graph

AllSci developed AERIS chat to dialogue with researchers in plain language while grounded in the scientific corpus — every claim traced back to the article, trial, hypothesis or patent it came from.

✦AERIS · grounded in 14 sources
Across recent trials, Rituximab shows a marked reduction in annualised relapse rate in RRMS, though head-to-head data vs. ocrelizumab is limited…
NEJM 2022NCT04*****Hypothesis #2,481
Agents

Agents that read the graph and support R&D

AllSci developed agents for specific scientific tasks and workflows such as literature reviews, competitive landscapes, asset assessments, and clinical trial designs.

Competitive landscape · anti-CD20 in MSRun ▸
  1. ✓Pull every trial, asset and program for the target
  2. ✓Cluster by mechanism, phase and sponsor
  3. ◔Draft a benchmarked summary with citations
AERIS · grounded in 214 sources
AI peer review

Structured review, instantly

AllSci developed an AI peer review to deliver structured, evidence-based checks of any new science — an instant first-pass assessment for quality signaling.

AI reviewReviewing “B-cell depletion lowers MS relapse rate”
Grounded in the cited evidence?Agree
Methodology sound?Agree
Novel vs. prior work?Partly
Every judgement traces back to the supporting and refuting evidence.
News generation

News generation

AllSci developed a data-driven news creation technology leveraging proprietary agents, workflows, and human editors.

FDA approves anti-CD20 therapy for RRMS — Phase III readout2h
Roche inks $1.5B deal for obesity asset5h
New hypothesis links CD36 to a metabolic checkpoint1d

Meet AERIS

AllSci Enhanced Research Intelligence System

AERIS breaks scientific publications down into hypotheses, research questions and results — grounded to their source, and searchable across one integrated graph.

Data Extraction→
Hypothesis Generation→
Validation→
Insights Sharing
Unstructured raw data
  • 276M+ Articles
  • 1.3M+ Clinical trials
  • 5.8M+ Patents
  • 1.6M+ Grants
AERIS AI

A family of 8B–70B parameter models, custom-tuned for scientific reasoning, search and extraction.

↓ Atomize
  • 17.7MQuestions
  • 17MHypotheses
  • 1.3MResults
Collection of atoms
  • Linked to source · grounded generation
  • Explainable reasoning & transparency
  • Interlinked knowledge graph
  • Embeddings + graph-DB powered search