The Problem: Manual Clinical Coding in GP Practices
Every day, NHS GP practices receive a steady stream of medical documents — pathology results, radiology reports, discharge summaries, and clinical letters. Each one needs to be:
- Read and understood — often from scanned PDFs or inconsistent formats
- Patient-matched — the correct patient identified in the GP system
- Clinically coded — findings mapped to SNOMED-CT codes
- Filed — coded data entered into the GP clinical record
For most practices, this is done manually. A clinician or clinical administrator reads the document, looks up the patient on PDS, searches for the right SNOMED codes, and types (or copy-pastes) the information into the GP system.
It takes roughly 15 minutes per document. Multiply that across the hundreds of documents a busy practice processes daily, and you're looking at hours of clinician and administrative time consumed by repetitive, error-prone data entry.
Our Solution: GPCopilot
GPCopilot is a multi-agent AI system that automates this entire workflow. Built on the Microsoft Agent Framework and integrated with core NHS FHIR APIs, it handles the heavy lifting while keeping clinicians in the loop.
How It Works
GPCopilot runs a five-step automated pipeline:
Step 1: Extract — Our document extraction agent parses the uploaded document (PDF, scan, or MESH message) and extracts structured clinical findings, patient demographics, and metadata.
Step 2: Identify — The patient identification agent matches extracted demographics against the NHS Personal Demographics Service using FHIR R4 APIs. Smart matching uses NHS number, name, date of birth, and postcode.
Step 3: Code — The clinical coding agent maps each finding to SNOMED-CT codes via the NHS Terminology Server. Every suggestion comes with a confidence score and alternative options.
Step 4: Validate — This is the human-in-the-loop gate. Clinicians review the coded findings in a clear dashboard. They can approve, modify, or reject any code. High-confidence codes can optionally be auto-approved based on practice policy.
Step 5: Update — Approved codes are filed to the GP clinical system via GP Connect Send Document and MESH. A full audit trail is maintained from upload to filing.
Built on NHS Standards
GPCopilot isn't a black box that sits outside your NHS infrastructure. It integrates natively with:
- NHS PDS (FHIR R4) for patient identification
- NHS Terminology Server (FHIR R4) for SNOMED-CT coding
- GP Connect (FHIR STU3) for record access and document filing
- MESH for secure inter-organisation messaging
All data stays within UK Azure regions. All connections use NHS-standard security (mTLS for Spine, signed JWT assertions for API auth).
Why We Built This
The NHS is under more pressure than ever. Clinicians shouldn't be spending their expertise on manual data entry when AI can handle the extraction and coding — with clinician oversight at every step.
We built GPCopilot because we believe:
- AI should assist, not replace — clinicians review every decision
- NHS standards matter — we use FHIR, SNOMED-CT, and NHS APIs, not proprietary workarounds
- Clinical safety comes first — we follow DCB0129/DCB0160 and maintain full audit trails
- Time saved is care gained — every hour saved on coding is an hour returned to patient care
What's Next
We're currently running pilot programmes with GP practices. If you're interested in seeing GPCopilot in action, request a demo — we'd love to show you what's possible.
GPCopilot is a product of Kinetix AI labs Ltd. For more information, visit kinetix-labs.ai or email hello@kinetix-labs.ai.