Kyber Clinical

AI-Powered Clinical Trial Intelligence Platform

Investor Presentation • May 2026 • Confidential

The Problem

A single Phase II trial generates 3,000-5,000 documents across 48+ TMF sections, managed across multiple disconnected systems. Nobody has a complete picture.

$4.2M
Average cost of a protocol deviation

Each deviation triggers rework, queries, potential regulatory findings, and timeline delays. Most deviations stem from document management failures — missing signatures, outdated versions, missed filings.

68%
Of TMFs fail regulatory inspection

The Trial Master File is the single source of evidence for regulators. Two-thirds fail on first inspection — missing documents, incomplete filing, version control failures, unsigned forms.

40%
Of CRO time spent on admin, not science

Clinical research associates spend nearly half their time on manual document tracking, status chasing across systems, duplicate data entry, and filing — not on patient safety or data quality.

Root cause: Clinical trial data lives in silos — Veeva for documents, Medidata for EDC, SharePoint for local files, SAS datasets for SDTM/ADaM submissions. No single system understands all of it. Nobody is connecting the dots.

Clinical Trial Data Standards

Clinical trials operate across multiple data standards, each serving a different purpose. No single platform today connects all three layers. That's the gap Kyber fills.

SDTM — Study Data Tabulation Model

What it is: The CDISC standard for structuring raw clinical trial data for regulatory submission. Every patient visit, lab result, adverse event mapped to standard domains (DM, AE, LB, VS, CM).

Who requires it: FDA, EMA, MHRA — mandatory for drug approval submissions.

Format: Static tabular datasets (SAS transport files / CSV). End-of-study. Not real-time. Not interoperable.

ADaM — Analysis Data Model

What it is: Analysis-ready datasets built on top of SDTM. Adds derived variables, statistical metadata, and analysis groupings so statisticians can run analyses without restructuring.

Who requires it: Also mandatory for regulatory submissions alongside SDTM.

Format: Static datasets. Used at study end for statistical analysis and submission. Contains the "truth" that regulators verify against.

FHIR — Fast Healthcare Interoperability Resources

What it is: HL7's modern standard for real-time healthcare data exchange. REST API, JSON payloads, standard resource types (Patient, Observation, ResearchStudy, DocumentReference).

Who's adopting it: NHS England, MHRA, FDA, Epic, Cerner, Veeva, Medidata. Becoming mandatory for NHS interoperability by 2027.

Format: Live, real-time API. How systems talk to each other during a trial — not just at the end.

The Integration Gap

Today's clinical trial stack has three disconnected data layers. Nobody connects them. Kyber does.

Regulatory Submission Layer — SDTM / ADaM

Tabular datasets submitted to FDA/EMA at study end. Structured, validated, static. SAS datasets, define.xml. Nobody looks at these during the trial — they're built after database lock.

Operational Intelligence Layer — THIS IS KYBER

TMF document intelligence, rules engine, AI copilot, audit trail. Reads documents (TMF), ingests datasets (SDTM/ADaM), communicates via FHIR with live systems. Single source of truth across all three layers. The brain that connects everything.

Live Clinical Systems — FHIR

Hospital EHRs (Epic, Cerner), EDC platforms (Medidata Rave, Oracle InForm), eTMF systems (Veeva Vault), CTMS, IWRS. Real-time data flowing during the trial. FHIR is how they talk to each other.

Why this is our moat: Carelane and other competitors are trying to replace the bottom layer (be the EDC). We're building the middle layer that connects everything. Different buyer, different sales cycle, much harder to displace once installed. Every system that connects to us makes us more valuable — network effects.

How FHIR Mapping Works

Everything in a clinical trial maps to a FHIR resource. This is how Kyber creates a single source of truth — regardless of which system the data came from.

Clinical Trial Concept → FHIR Resource

A clinical trial (study protocol)
ResearchStudy
An enrolled patient (subject)
ResearchSubject
A TMF document (any filing)
DocumentReference
A protocol deviation
DetectedIssue
A CAPA (corrective action)
Task
A study site (hospital/clinic)
Location
An adverse event (SAE/AE)
AdverseEvent

Why this matters practically

Same query, any source system

Once everything is a FHIR resource, you can query "show me all SAEs at Site 003" regardless of whether the data came from Veeva, Medidata, or a spreadsheet upload. The query is the same because the data model is the same.

AI reasons over standard resources

The AI copilot doesn't need to understand Veeva's schema AND Medidata's schema AND SharePoint's folder structure. It queries FHIR resources. One data model to learn, infinite sources to connect to.

Regulators speak FHIR too

FDA, EMA, and MHRA are all moving toward FHIR-based submissions. When our data is already FHIR-native, regulatory readiness becomes a byproduct of how we store data — not a separate submission exercise.

Example: Study → ResearchStudy

DEOX-MOCK-002, our mock study, mapped to a FHIR ResearchStudy resource. This is what the JSON actually looks like.

What exists in the database today:

// Current Kyber internal data
{
  "study_id": "DEOX-MOCK-002",
  "study_name": "NOVA-STAR",
  "phase": "Phase II",
  "indication": "NSCLC, EGFR-mutant",
  "drug": "Bica-Xi (bispecific antibody)",
  "sites": 6,
  "enrolled": 47,
  "target_enrollment": 120,
  "tmf_docs": 318
}

Mapped to FHIR ResearchStudy resource:

{
  "resourceType": "ResearchStudy",
  "id": "DEOX-MOCK-002",
  "title": "NOVA-STAR",
  "status": "active",
  "phase": {
    "text": "Phase II"
  },
  "condition": [{
    "text": "NSCLC, EGFR-mutant"
  }],
  "description": "Bica-Xi bispecific antibody",
  "enrollment": [{
    "reference": "Group/DEOX-MOCK-002-enrolled"
  }]
}

What this enables: Any FHIR-compliant system can now query this study. The AI copilot can say "Study DEOX-MOCK-002 is Phase II, NSCLC, 47 of 120 enrolled" by querying a single resource type — regardless of whether the data came from our database, Veeva, or a spreadsheet.

Example: TMF Document → DocumentReference

Every document filed in the Trial Master File maps to a FHIR DocumentReference. This is how we track 524 artifact types across 11 zones.

TMF filing in the database:

// A signed informed consent form
{
  "doc_id": "DOC-20260420-a3f2k",
  "study_id": "DEOX-MOCK-002",
  "title": "ICF_v3.0_Signed_2026-04-20.pdf",
  "zone": "04 IRB/IEC",
  "section": "04.02 Consent",
  "artifact": "04.02.02",
  "status": "filed",
  "site_id": "SITE-103",
  "uploaded_by": "jane.smith@cro.com",
  "r2_key": "DEOX-MOCK-002/tmf/04/04.02/..."
}

Mapped to FHIR DocumentReference:

{
  "resourceType": "DocumentReference",
  "id": "DOC-20260420-a3f2k",
  "status": "current",
  "type": {
    "text": "Informed Consent Form",
    "coding": [{
      "code": "04.02.02",
      "system": "dia-tmf-v3.1"
    }]
  },
  "subject": {
    "reference": "ResearchStudy/DEOX-MOCK-002"
  },
  "date": "2026-04-20T14:32:00Z",
  "content": [{
    "attachment": {
      "url": "r2://DEOX-MOCK-002/tmf/04/...",
      "contentType": "application/pdf"
    }
  }]
}

What this enables: The TMF classifier can auto-classify uploaded documents against the DIA TMF Reference Model by setting the type.coding.code. The rules engine can query "how many DocumentReferences exist for zone 04?" to calculate TMF completeness. The copilot can answer "what's missing for inspection readiness?" by comparing filed DocumentReferences against the expected 524 artifact types.

Example: Protocol Deviation → DetectedIssue

Protocol deviations, CAPAs, and adverse events all map to FHIR resources. Here's how a deviation flows through the system.

Deviation record:

{
  "id": "DEV-2026-0047",
  "study_id": "DEOX-MOCK-002",
  "site_id": "SITE-103",
  "type": "Informed Consent",
  "severity": "major",
  "description": "ICF v3.0 used after v3.1
                   approved by IRB",
  "subject_id": "SUBJ-103-012",
  "detected_date": "2026-05-15",
  "capa_id": "CAPA-2026-0018"
}

Mapped to FHIR DetectedIssue + linked Task:

{
  "resourceType": "DetectedIssue",
  "id": "DEV-2026-0047",
  "status": "preliminary",
  "severity": "high",
  "code": {
    "text": "Protocol Deviation"
  },
  "subject": {
    "reference": "ResearchSubject/SUBJ-103-012"
  },
  "detail": "ICF v3.0 used after v3.1...",
  "mitigation": [{
    "action": {
      "reference": {
        "reference": "Task/CAPA-2026-0018"
      }
    }
  }]
}

The chain of linked resources: ResearchStudy → Location (Site-103) → ResearchSubject (patient) → DetectedIssue (deviation) → Task (CAPA). Every resource links to the others. The AI copilot can traverse this chain: "Subject 103-012 had a major informed consent deviation on May 15th, CAPA-0018 is assigned to resolve it." All from standard FHIR queries — no custom schema needed.

What Kyber Actually Does

Kyber is the intelligence layer that sits on top of existing clinical trial infrastructure — reading documents, classifying data, enforcing rules, and providing AI-powered insights across the entire trial.

TMF Document Intelligence

Auto-classifies documents against the DIA TMF Reference Model (524 artifact types across 11 zones). Detects missing filings, version conflicts, unsigned documents. Flags inspection readiness gaps before regulators find them.

Clinical Rules Engine

12 configurable clinical rules with adjustable thresholds. Monitors safety signals, enrollment velocity, deviation patterns, site performance, query aging, TMF completeness. Fires real-time alerts when rules are breached.

AI Copilot

Context-aware clinical AI assistant. Ask "How many SAEs at Site 003?" or "What TMF gaps for inspection readiness?" Classifies intent across 17 clinical modules. Conversation memory persists across sessions.

FHIR-Native Data Model

All data normalised to HL7 FHIR resources. Ingest from any FHIR-speaking system — Veeva, Medidata, Epic. Aligned with NHS England and MHRA interoperability mandates for 2027.

What We Have Today

Working platform with mock study data, API backend, and clinical domain scaffolding. Honest assessment.

Working Now

  • 26-tab dark UI with mock study (NOVA-STAR: Phase II NSCLC, 6 UK sites, 47/120 enrolled)
  • 100+ API route handlers on Cloudflare Workers
  • D1 database with CRUD, R2 for document storage
  • TMF Reference Model: 524 artifacts, 11 zones, 48 sections
  • Rules engine: 12 clinical rules evaluating (28KB)
  • AI Copilot wired to GLM with dual-key rotation
  • 17 module-specific system prompts
  • Git version-controlled, CI/CD via Cloudflare Pages

Shipping Next 4 Weeks

  • Copilot fully live with persistent memory
  • Study Brain: contextual AI queries
  • Rules engine wired to API (real-time alerts)
  • TMF auto-classification (upload → classify)
  • Document generation (real Office output)
  • Audit trail (real tracking)
  • SharePoint write-back (Graph API)
  • FHIR resource layer for mock study
  • Real auth with sessions and roles

Later (requires funding)

  • Veeva Vault connector
  • Medidata Rave connector
  • SDTM/ADaM dataset ingestion
  • FHIR API gateway for live systems
  • OCR document extraction
  • Mobile-responsive redesign
  • GxP validation package
  • SOC 2 / ISO 27001
  • Multi-tenant provisioning

Build Roadmap

From current state to investor-demo competitive in 4 weeks. All built on existing scaffolding.

Week 1 — AI Core

BY MAY 28

  • Copilot live with GLM + memory
  • Study Brain connected to AI
  • 12 rules evaluating & firing
  • All 501 stubs replaced
  • Intent classification live

Week 2 — Documents

BY JUNE 4

  • TMF Classifier: upload → classify
  • Document Generator: real output
  • Audit Trail: real tracking
  • Algorithm weights in calculations
  • Real calculated metrics

Week 3 — Integration

BY JUNE 11

  • SharePoint write-back
  • FHIR resource layer
  • Copilot context-aware
  • Data flow across all tabs
  • Performance optimisation

Week 4 — Production

BY JUNE 18

  • Real auth: login, roles
  • Multi-study support
  • External demo-ready
  • Investor walkthrough
  • Production deploy

Competitive Landscape

Carelane — €3.4M raised, 170+ sites

Full EDC, eCOA, ePRO, consent, FHIR-native protocols, AI CRF builder, SOC 2/HIPAA/GDPR. Working AI agent.

Weakness: Rip-and-replace model. 18-month sales cycles. Enterprise pricing. No TMF intelligence focus.

Kyber — Intelligence Layer

Connects to existing Veeva/Medidata/SharePoint. TMF intelligence (524 artifacts). Rules engine (12 rules). AI copilot. FHIR-native data model.

Advantage: Days to deploy. No adoption friction. We sell to clinical ops, not IT.

Key insight

Carelane owns the entire clinical trial stack. We build the brain that connects it all. Different market, different buyer, different sales cycle. We sell to the Clinical Operations Director — not the CIO.

Market Opportunity

$8.5B
CTMS Market (2026)

Growing at 13.8% CAGR. Legacy platforms 15+ years old, lack AI.

$3.2B
eTMF Software Market (2026)

Veeva and MasterControl dominate. Designed for storage, not intelligence.

4,200+
UK Clinical Trials (2025)

MHRA tightening. NHS mandating FHIR by 2027. UK-first tailwinds.

eTMF Intelligence Document Classification Rules Engine AI Copilot FHIR Interoperability SDTM/ADaM Integration Inspection Readiness Regulatory Automation

Business Model

SaaS Platform

Per-study, per-site pricing. AI copilot, document intelligence, rules engine, FHIR data access.

Target: £2,000-5,000/study/month

No system replacement required. Onboarded in days.

Professional Services

Setup, training, validation. TMF migration, study setup, rules config, FHIR connector deployment.

Target: £15,000-30,000/implementation

Also: inspection readiness audits, TMF gap analysis.

IP Licensing

White-label the intelligence engine and rules engine to CTMS/eTMF vendors who lack AI.

Target: Revenue share or per-seat

Every CTMS vendor becomes a customer, not a competitor.

The Team

JT
Joe Thornton
Founder & CEO
Clinical Development Director. Deep expertise in clinical trial operations, QMS, CRO oversight, GCP. Previously Veeva Systems, Rivia Health, VITA FSP.
AI
AI Engineering Stack
Technical Co-founder
Cloudflare Workers + D1. Multi-model AI. Microsoft Graph API. FHIR resource model. 229 source files, 57K lines. Built by AI under Joe's direction.

Advisory Network

Career relationships with Veeva, Rivia, VITA FSP. Active network across UK CROs, pharma sponsors, regulatory consultants.

The Ask

Seeking Seed Investment

Complete the 4-week build sprint, secure first pilot customers, prove the intelligence layer model with real clinical trial data.

Use of Funds

  • 40% — AI Backend: Copilot, Study Brain, TMF Classifier, RAG, FHIR layer
  • 25% — Connectors: Veeva, Medidata, FHIR API gateway, SDTM/ADaM
  • 20% — Sales: First 3 pilot customers, CRO partnerships
  • 15% — Compliance: GxP validation, SOC 2, ISO 27001

12-Month Milestones

  • Intelligence layer live and demo-able
  • 3 pilot customers with real trial data
  • £100K ARR
  • Veeva or Medidata connector in beta
  • First TMF gap analysis as paid engagement
  • SDTM/ADaM ingestion for one study
  • FHIR API gateway with live source system
  • SOC 2 Type II started

Let's Build

The intelligence layer for clinical trials.

Joe Thornton, Founder
joe@kyberclinical.com
kyberclinical.com