KNOWLEDGE GRAPH · MEDICAL
DESIGN-PARTNER PIPELINE · BIRAC-TRACKEDHospital imaging, sequencing data, EHR archives, device-tracking logs, clinical-trial cohorts — every data type a regulated medical estate produces, compressed at the data plane and linked into one audit-trailed knowledge graph. HIPAA-aligned deployment patterns, DPDP Act 2023 residency by default, no public-LLM exposure.
01 — THE PROBLEM
A modern hospital generates more imaging in a year than every radiograph it produced in its first three decades. A single FASTQ run from a sequencing lab is 100–300 GB before alignment. An orthopaedic-implant manufacturer carries a 25-year follow-up obligation against every device shipped. A clinical-trial dataset is retained for the lifetime of the regulatory submission plus another decade.
Medical estates retain everything, forever, by regulation. The cost curve is unforgiving — and the conventional answer (“move it to the cheap cold tier”) breaks the day someone asks a clinical question that needs the data hot again. Storage costs compound. Imaging archives migrate every 5–7 years from one PACS vendor to another and lose provenance in transit. Sequencing labs throw away results because the storage bill is bigger than the experiment.
On top of which, the data is structurally siloed. Imaging sits in PACS. EHR sits in the hospital information system. Genomics sits in a lab LIMS. Device-tracking logs sit in the manufacturer's warranty database. Clinical-trial assays sit with the CRO. Every cross-system question — “how did this implant's outcome correlate with this patient's imaging history?” — costs a quarter and three analysts.
Public-LLM tools are non-starters. Patient data crosses regulatory, ethical, and contractual boundaries the moment it leaves the hospital perimeter. The medical estates that need this capability the most are the ones structurally blocked from the off-the-shelf product.
02 — WHAT IT COVERS
Cithorum's Medical KG is the unified data layer for the full clinical and research estate. The same compression substrate and the same provenance graph carry every data class.
SEQUENCING · FLAGSHIP
FASTQ, BAM, VCF — raw reads through aligned variants. 12% smaller-on-disk, 22-min alignment on the SwissVault genomics benchmark. Beyond the archive: a closed-loop reasoning layer for antimicrobial resistance (AMR) in tertiary hospitals — bind sequence → phenotype → mechanism → provenance into a single auditable clinical call.
IMAGING · FLAGSHIP
CT, MRI, PET, X-ray, ultrasound, DICOM. Petabyte-scale hospital archives compressed 3–8× live and up to 100× on cold-tier history. Imaging linked to patient, study, referring physician, and outcome — survives the next PACS migration with full provenance intact.
DEVICES
Device serial → patient → procedure → surgeon → long-term outcome → manufacturer → recall graph. The missing layer between an orthopaedic implant's 25-year follow-up obligation and the data that proves or breaks the warranty claim.
EHR · CLINICAL
Structured records and unstructured notes archived with per-record encryption. Long-retention compression on de-active records. Patient → visit → diagnosis → medication → provider graph.
TRIALS
Cohort data, raw assay outputs, inter-site replicates, statistical packages. Retained for the lifetime of the submission plus regulator-mandated decades. Trial → site → patient → assay → outcome → submission graph.
PROCUREMENT
Order → manufacturer → distributor → hospital → department → procedure consumption. Where a medical-device distributor sees what their devices actually do once they enter clinical use — at population scale.
02A — PLATFORM
Cithorum sits on top of the hospital's existing systems-of-record — LIMS, PACS, EMR, sequencing instruments — and emits queryable, audit-trailed intelligence back out to the clinical decision surfaces. We don't replace the systems of record. We bind the evidence across them and give the analyst a graph to reason over, not another dashboard to ignore.
INGEST
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.
REASONING LAYER
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
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.
(p:Patient)-[:IN]->(w:Ward)<-[:IN]-(q:Patient)
-[:CULTURED]->(i:Isolate {genus:"Klebsiella"})
WHERE q.admit < p.admit AND p.admit < q.discharge(:Isolate)-[:CARRIES]->(:Gene {name:"bla_KPC"})
-[:ON]->(:Plasmid {type:"IncFII"})(d:Device {sn:"SN-83"})-[:IMPLANTED_IN]->(p:Patient)
-[:UNDERWENT]->(s:Surgery)
-[:OUTCOME]->(o:Outcome) WHERE o.years_post >= 5Not 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.
02C — APPS ON THE MEDICAL KG
The reasoning layer is the architecture; the apps are how clinicians, lab directors, and quality leads actually use it. Each app is a productised work surface for a specific clinical or lab workflow, running on the same KG substrate, the same Jam-compressed storage, the same regulatory posture. Buy one app to start; add the rest as your estate matures.
The sequencing-lab work surface. FASTQ ingest with Jam compression at the data plane, hash-verified batch management, variant interpretation workflow, regulator-export PDFs. Built on the SwissVault genomics benchmark — 12% smaller-on-disk, 22-min alignment.
Buyers: sequencing labs, biobank custodians, clinical genomics units.
Closed-loop clinical AMR decisioning. Predict → Measure → Confirm → Retrain over CRE-Kp and the broader carbapenem-resistant Enterobacterales. Try the live walkthrough on this page; the productised version ships Q4 2026 with the first design-partner hospital.
Buyers: tertiary-care hospital CIOs, infection-control leads, clinical microbiology directors.
Radiology worklist with KG context. Prior studies, ward, referring physician, outcome, and similar-case retrieval surfaced inline at the read. PACS-integration ready (DICOM + HL7), customer-controlled deployment.
Buyers: radiology IT leads, imaging-archive custodians, hospital CMIOs.
Implant serial → patient → procedure → surgeon → 5-year outcome → manufacturer → recall graph. The post-market surveillance app for orthopaedic and other long-follow-up device classes. Anchor: warm intro via Innoways Healthcare network.
Buyers: device-manufacturer quality & compliance, regulatory affairs, distributor networks.
Multi-site cohort + assay tracking. Inter-site replicate management, regulatory submission export, decade-plus retention with Jam-compressed archive economics. Trial → site → patient → assay → outcome → submission graph.
Buyers: trial sponsors, CROs, biometrics leads.
What devices actually do once they enter clinical use, at population scale. The manufacturer / distributor view — order → hospital → department → procedure → outcome. Reads the device-tracking graph; produces the commercial signal.
Buyers: device manufacturers, distributors, hospital procurement leads.
Buy one app, add the rest as your estate matures. Each app inherits the platform's sovereign deployment, HIPAA / DPDP / ICMR posture, no-public-LLM guarantee, and audit-trailed event log. Multi-app sites get bundle pricing. Every app is removable without touching the systems of record — the LIMS, PACS, and EMR keep running.
02B — FLAGSHIP CLINICAL DOMAIN
The sharpest medical use-case for the Cithorum substrate is the one with the strongest scientific consensus and the biggest clinical urgency: antimicrobial resistance in Indian tertiary care — and specifically the carbapenem-resistant K. pneumoniae (CRE-Kp) loop. ICMR 2022 reports 56% of K. pneumoniae isolates meropenem-resistant in tertiary sentinel sites. The diagnostic stack (VITEK, BD Phoenix, MALDI-TOF) produces results; what's missing is the reasoning layer that binds the result to ward action in time.
The open quadrant. The clinical AMR decision layer is structurally vacant. Pathogenwatch and amr.watch are population-level epidemiology, not point-of-care clinical. VITEK / BD Phoenix / Accelerate are wet-lab hardware — the input layer to the loop, not the reasoner. Causaly, BenevolentAI, Recursion, Aganitha are pharma R&D graphs, off-grid for clinical workflow. LIMS systems (LabWare, STARLIMS) are the substrate the loop writes to and reads from — integration surface, not competitor. Cithorum's Medical KG sits squarely in the open software-clinical quadrant, designed for the Indian tertiary-care workflow as the design target rather than the export.
SCIENTIFIC SUPPORT
Pitout JDD, Peirano G, Kock MM, et al. The Global Ascendency of OXA-48-Type Carbapenemases. Clinical Microbiology Reviews 2019; 33(1):e00102-19.
The KG models plasmids and clones as first-class entities — what the diagnostics and surveillance layers cannot model on their own.
SCIENTIFIC SUPPORT
Chen et al. 2023. Journal of Hospital Infection. DOI 10.1016/j.jhin.2023.07.002.
The KG carries ward and prior-occupant context end-to-end so the loop can flag colonisation risk before the next admission.
SCIENTIFIC SUPPORT
Tamma PD, Aitken SL, Bonomo RA, et al. Clinical Infectious Diseases (IDSA Guidance) 2022; 74(12):2089-2114.
The closed loop binds prediction to confirmed AST — empirical or ML-only prediction is not a substitute for phenotypic confirmation.
TARGET MARKET
Top-10 private chains (Apollo, Fortis, Manipal, Max, Medanta, Narayana, Kokilaben, Artemis, Wockhardt, AIG) plus ~30 large tertiary government / academic centres (AIIMS network, PGIMER, JIPMER, CMC Vellore, KEM). Site-level licence economics map cleanly onto the BIRAC-tracked biomedical instrument.
03 — HOW IT WORKS
01
Jam at the data plane. 3–8× live compression across imaging and EHR; up to 100× on cold-tier and backup workloads. NVMe-bound throughout. The archive bill stops being the constraint.
02
Typed entities (patient, study, variant, device, trial, outcome) with typed edges (linked-to, replaces, resulted-in, recalled-by, prescribed-by). Schema versioned per regulatory framework. Provenance traced end to end.
03
Ask the graph in natural language. The AI reads the schema; it never sees raw imaging or genomic content unless explicitly permissioned. Every model call audit-logged for regulator review.
04 — REGULATORY POSTURE
Deployment patterns built for HIPAA Privacy and Security Rule control families. Per-tenant encryption, per-edge permissions, audit-trailed access. Designed to map cleanly to BAA obligations.
Sovereign-by-default Indian deployment via Cithorum Cloud. Patient data stays in-India, in the customer's tenancy. CERT-In incident-response logging integrated. No cross-border transfer without explicit clinical authorisation.
Indian Council of Medical Research alignment for research-data handling. Genomic data biobank policies, consent provenance per record, lifecycle policies for trial data retention. Audit-export ready.
AI runs on customer-deployed or local models. Patient content never leaves the perimeter. The model sees the schema, not the bytes. Air-gapped option for the most sensitive estates.
Runs on infrastructure the customer owns or controls. On-premise for hospital systems, sovereign cloud via Cithorum Cloud for distributed networks, air-gapped for clinical-trial sites.
Jam compression turns regulator-mandated retention from a 25-year cost commitment into a manageable line item. Per-TB economics improve year over year as compression ratios mature.
05 — WHO BUYS IT
CIOs, CMIOs, radiology IT leads.
Imaging archive growth is the canonical pain point. Cithorum collapses the PACS bill, unifies imaging with EHR, and gives the AI assistant a graph to reason over without leaving the perimeter.
Lab directors, bioinformatics leads, biobank custodians.
FASTQ-class compression is where Jam compounds hardest. The 12% smaller-on-disk + 22-min alignment delta plus full chain-of-custody makes the cost story and the regulatory story one and the same.
Quality & compliance officers, regulatory affairs leads.
Implant and device makers carry decade-plus follow-up obligations. The Medical KG turns serial numbers + procedures + outcomes into a queryable graph — recall management, post-market surveillance, long-term efficacy claims.
Trial operations, data managers, biometrics leads.
Multi-site trial data, raw assays, inter-site replicates, regulatory submissions. The graph models the full data lineage; the substrate keeps the long-retention bill survivable.
06 — STATUS
Cithorum is actively engaging design partners across hospital systems, sequencing labs, and medical-device manufacturers in India and Europe. The biomedical workstream is led by Suraj S Naik (Co-founder · Head of Government Procurement and Tender Relationships; Chirag Labs) and is being tracked against the BIRAC ₹24.37 Cr non-dilutive biomedical instrument. The SwissVault genomics benchmark trial confirms the FASTQ-class compounding. Six-month paid design-partner engagements available; named technical contact assigned per partner.
Apply for design partnership →