Knowledge Graph Platforms

Customer data becomes queryable. Raw bytes stay private.

Cithorum builds vertical Knowledge Graph platforms over regulated data. Every fact has provenance. AI sees the graph schema and permitted facts, not a dump of customer records into public models.

Discuss a design partnership →Open KG explainer →

Medical pipeline · platform partners · graph proof

SwissVault genomics

01/Shared substrate

One graph core. Different vertical schemas.

The same primitives repeat across Data-Centre Ops, Medical, Law, and Workspace: typed entities, typed edges, permissioned facts, and a full audit chain.

3External verticalsMedical · Law · Workspace
100%Audit-trailedProvenance per fact
0Public-LLM exposureSchema-only graph access
1Medical grant routeOptional non-dilutive path

Entities

Patients, matters, accounts, workloads, nodes, documents, devices.

Edges

Relationships, ownership, provenance, lineage, custody, decisions.

Permissions

Per-tenant and per-fact controls before any AI surface sees data.

02/Interactive map

Same core, different graph shape.

Select a vertical to see which entities it reads, which it writes, and how the shared substrate changes per buyer.

03/Data-Centre Ops KGInternal alpha

The first graph runs the cloud.

Cloud Ops KG links customers, buckets, workloads, nodes, drives, restore requests, power draw, and billing facts. It is closing as the internal proof layer for Cithorum Cloud and the pattern for external KG platforms.

Customer view

What data exists, where it lives, what it costs, and when it restored.

Operator view

Which node, drive, service, actor, or workload changed each fact.

AI boundary

Assistants can query permitted graph facts without seeing raw customer bytes.

04/Medical KGDesign-partner pipeline

Medical data, linked without leaving the estate.

PACS, sequencing, EHR, devices, and clinical-trial records become one audit-trailed graph. A medical grant route remains viable for this vertical, but it is not the main Cithorum raise.

12%Smaller FASTQSwissVault genomics
22 minutesAlignment pathSwissVault benchmark
0Public model exposureNo raw medical bytes
100%ProvenanceEvery fact tied to source

Cithorum sits on top of hospital systems of record: LIMS, PACS, EMR, sequencing instruments, and device systems. It does not replace them. It makes their facts queryable.

SOURCESCITHORUM KG COREQUERY & ALERT SURFACE

INGEST

What Cithorum reads from

Cithorum sits on top of the hospital's existing systems-of-record. We don't replace the LIMS, the PACS, or the EMR — we read their output, normalise it, and bind it into the graph with full provenance.

  • 01PACS · DICOM imaging archives (CT, MRI, X-ray, ultrasound)
  • 02LIMS · LabWare, STARLIMS, Clarity (sample → result records)
  • 03EMR · admissions, orders, dispensing, discharge
  • 04Sequencing instruments · short-read WGS, MALDI-TOF, plasmid typing
  • 05Diagnostics · VITEK 2 / BD Phoenix antibiograms, MIC values
  • 06Device registries · implant serial → procedure → outcome

REASONING LAYER

The typed knowledge graph

IsolatePatientWardGenePlasmidDrugOutcomeMIC

8 node types · 14 edge types · typed schema v0.4

Every edge carries provenance: source, timestamp, actor. Every fact traces. The graph is the integration surface, not a silo on top of one.

QUERY & ALERT

What Cithorum writes back out

Queries — not dashboards. Alert surfaces are typed, not preset. The graph emits answers wherever they need to land: at the bedside, in the clinician's inbox, in the regulator's audit report.

WARD CO-COLONISATION
(p:Patient)-[:IN]->(w:Ward)<-[:IN]-(q:Patient)
-[:CULTURED]->(i:Isolate {genus:"Klebsiella"})
WHERE q.admit < p.admit AND p.admit < q.discharge
Ward co-colonisation chain · window 0–30d
PLASMID-LEVEL RESISTANCE
(:Isolate)-[:CARRIES]->(:Gene {name:"bla_KPC"})
-[:ON]->(:Plasmid {type:"IncFII"})
Plasmid-level resistance shell
IMPLANT POST-MARKET
(d:Device {sn:"SN-83"})-[:IMPLANTED_IN]->(p:Patient)
-[:UNDERWENT]->(s:Surgery)
-[:OUTCOME]->(o:Outcome) WHERE o.years_post >= 5
Long-term implant outcome trace
  • Ward dashboardsRisk-tiered isolate alerts at the bedside
  • Clinician inboxDiscordance flags, escalation prompts
  • Audit & regulator reportsProvenance-traced, AST-confirmed
  • AI assistantSchema-only — never raw PHI or imaging

Not another LIMS. LabWare, STARLIMS, Clarity, and the hospital's PACS / EMR / HIS remain the systems of record. Cithorum is the reasoning layer that reads from them, binds the evidence, and emits queryable, audit-trailed intelligence — without any data leaving the operator boundary.

WALKTHROUGHIllustrative — synthetic isolate, demo dataCITHORUM / ISO-004113 / PREDICT

05/Law KGDesign-partner gated

Matter memory with an audit trail.

Legal teams need precedent, evidence, drafts, emails, and advice histories to be searchable without leaking client material into a public model.

Matters

Clients, counterparties, issues, evidence, deadlines, authorities.

AI boundary

The model queries permissioned facts; it does not ingest the file room.

Auditability

Every answer can cite the matter source, actor, and timestamp.

06/Workspace KGDesign partner active

Revenue operations as a graph.

Workspace KG links accounts, people, projects, documents, decisions, finance, and communications so operators can ask questions that span tools.

Bring one vertical into focus.

Talk to us →Open KG resources →