LOT 5 — Research, Innovation and Development
BritiAI response — NHS SBS Healthcare AI Solutions Open Framework (SBS10523)
Lot 5 rewards genuine research capability — published, benchmarked, and collaborative. BritiAI’s Lot 5 offer is anchored to the Beever AI research laboratory and the Votee MAGIC training platform, which together give this response a property unusual in the NHS AI vendor market: every methodological claim is supported by a peer-reviewed citation.
5.1 Why BritiAI wins Lot 5
Most AI suppliers presenting research credentials to the NHS are doing so for the first time. They have a product team, a sales team, and a marketing team — and a “research” function that, on inspection, is a content pipeline.
BritiAI’s research credibility sits in Beever AI, the Votee research laboratory, which operates as a working academic group with active collaborations at HKUST, the Vector Institute, Stanford, the University of Waterloo and HKU. The Beever laboratory has peer-reviewed publications at ACL 2025 — TheoremExplainAgent and HKCanto-Eval / CoNLL — and maintains the open-source Beever Atlas neural memory layer with a documented adopter community. The MAGIC large-language-model training platform that underpins our Lot 5 sovereign-training proposition is in production at enterprise and government clients and received the Baidu Innovation Award 2026.
In short: BritiAI is not learning the research process for the first time on NHS time. We are bringing an existing research operation to bear on NHS research priorities, and we are happy to be assessed on the published record.
5.2 Named solutions under Lot 5
(i) MAGIC-NHS — Sovereign LLM Training Pipeline
MAGIC-NHS is a managed environment that lets an NHS Approved Organisation fine-tune a large language model on its own clinical, operational and research data without that data leaving the organisation’s network boundary. The pipeline is the Votee MAGIC platform — a 95%-agentic LLM training stack — wrapped in the BritiAI delivery model and a Clinical Safety Case aligned to DCB0129.
A MAGIC-NHS engagement delivers, at minimum: a written data preparation and curation protocol; a training run on Approved Organisation-controlled infrastructure (on-premises or UK-sovereign private cloud); an evaluation report against published benchmarks relevant to the use case; a governance pack including a DPIA, a model card written to the emerging NHS England model-card standard, and a Clinical Safety Case; and a hand-over to the Approved Organisation’s own data science or digital function under the Lot 7 advisory wrap if required.
MAGIC-NHS is designed for Trusts, Integrated Care Boards and academic NHS partners that want a sovereign clinical or operational model — for example, a Trust-specific summarisation model trained on local clinical letters, or an ICB-level operational model trained on board papers, performance reports and policy. It is not designed to produce a diagnostic or treatment-recommending system; that boundary is set in writing in every MAGIC-NHS engagement.
(ii) Trial Concierge — Agentic Clinical Trial Recruitment and Protocol Support
Trial Concierge is an agentic system that supports clinical trial teams across three workflows. First, eligibility screening: the agent consumes structured and unstructured EHR signals against a documented protocol, surfaces candidate participants to the trial team with full reasoning trace, and never auto-enrols — clinical sign-off is mandatory at every step. Second, protocol Q&A: clinicians and trial co-ordinators can interrogate the live protocol in natural language and receive cited answers, dramatically reducing the response time on protocol queries that today flow through email. Third, recruitment letter drafting: the agent drafts patient-appropriate recruitment correspondence under the responsible clinician’s sign-off.
Trial Concierge is designed to plug into the CRF and EDC systems trial teams already operate, not to replace them. It does not make clinical eligibility determinations and does not modify the protocol. Where deployed alongside the NIHR portfolio or commercial sponsor trials, BritiAI will support the Approved Organisation’s R&D office through the relevant HRA and ethics engagements.
(iii) Cross-Trust Federated Research Substrate
Cross-Trust Federated Research Substrate deploys the Beever Atlas neural memory layer in a federated topology across multiple participating Approved Organisations. Each Trust’s underlying data remains in place, inside its own boundary; the substrate enables collaborative research workspaces — shared cohort definitions, shared evaluation protocols, shared model artefacts — without the data egress that today blocks most multi-Trust collaborations at the IG gate.
The substrate is built to be compatible with the Health Research Authority approvals process and with the NHS Research Ethics Committee model. BritiAI will support Trust R&D offices through the relevant approvals on a per-engagement basis under the Lot 7 advisory wrap.
5.3 Evidence base
- ACL 2025 publications from the Beever laboratory: TheoremExplainAgent (agentic reasoning) and HKCanto-Eval (low-resource language evaluation methodology) — both directly relevant to the agentic and evaluation methods used in Trial Concierge and MAGIC-NHS.
- Open-source release of Beever Atlas, with documented adopter community and benchmarked retrieval performance.
- MAGIC training platform in production at enterprise and public-sector clients; Baidu Innovation Award 2026; HKMA FSS 3.1 pilot programme.
- Active university partnerships: HKUST, Vector Institute, Stanford, University of Waterloo, HKU.
- BritiAI delivery infrastructure: the seven productised BritiAI capabilities (Social Listening, Copilot, ASR, Image Detection, Document Processing, Report Generation & BI, Data Processor) provide a documented engineering base into which research outputs can be productised under any Lot 5 call-off.
5.4 Delivery model
Lot 5 engagements are delivered against the BritiAI four-stage model — Diagnose, Design, Deploy, Demonstrate — with two Lot 5-specific adjustments. First, the Design phase explicitly produces a research protocol, an evaluation plan against published benchmarks, and a publication strategy where the Approved Organisation wishes to publish. Second, the Demonstrate phase produces a model card, a benchmark report, and (where applicable) a peer-reviewable methodology write-up suitable for submission alongside the Approved Organisation’s own academic outputs.
Every Lot 5 engagement is led by a named BritiAI principal and supported by Beever laboratory research staff for the duration. Beever is contracted to BritiAI under the framework Teaming Agreement, which flows down every UK GDPR, DSPT and DCB0129 obligation imposed by the Approved Organisation.
5.5 Outcome commitments
Lot 5 outcomes are inherently bespoke to the research question. BritiAI commits to the following outcome posture under any Lot 5 call-off:
- A written research protocol and evaluation plan agreed with the Approved Organisation’s R&D office before any model training begins.
- A model card and benchmark report delivered at the end of every MAGIC-NHS engagement.
- For Trial Concierge: a measurable uplift against the Approved Organisation’s baseline recruitment screening time and protocol-query response time, with the baseline and target agreed in writing during Diagnose.
- For the Federated Research Substrate: a participating-Trust-count and shared-cohort-count commitment agreed in writing per programme.
- No clinical or operational deployment of a research artefact without an accompanying Clinical Safety Case where DCB0129 applies.
5.6 Information governance, clinical safety and data sovereignty
Lot 5 work routinely operates on identifiable or pseudonymised data and therefore sits under the strictest BritiAI governance posture. Every Lot 5 engagement is operated under:
- UK data residency by default, with on-premises training supported for MAGIC-NHS where the Approved Organisation requires it.
- A DPIA and a Section 251 / Confidentiality Advisory Group engagement where the data flow requires it, supported by BritiAI’s information governance lead.
- DCB0129-aligned Clinical Safety Case where the research artefact is intended for clinical or operational deployment.
- HRA and REC engagement where applicable, with BritiAI supporting the Approved Organisation’s R&D office under the Lot 7 advisory wrap.
- No use of Approved Organisation data to train BritiAI, Votee or Beever foundation models without the Approved Organisation’s explicit written instruction.
5.7 Why this answer is different
Five points for the Lot 5 evaluator.
Sovereign by design. MAGIC-NHS lets the Approved Organisation fine-tune a model on its own data without that data leaving its network. There is no foreign sub-processing and no covert model-training reuse.
Clinically safe without overclaiming. Lot 5 research outputs are scoped explicitly outside Class IIa medical device territory at this round. Where a future research output requires a medical-device claim, BritiAI will follow the MHRA software-as-a-medical-device route in full before commencing.
Evidence-grade. Beever’s ACL 2025 publications and the open-source Beever Atlas release give every methodological claim in this response a citation. The Votee MAGIC platform is in regulated-sector production with documented enterprise references.
Embedded, not delivered. BritiAI does not parachute in a research team and depart. Every Lot 5 engagement is paired with a named principal who remains accountable through Demonstrate, with an optional Lot 7 wrap to support adoption and HRA engagement.
Buy-once, deploy-many. MAGIC-NHS, Trial Concierge and the Federated Research Substrate are productised engagement patterns, not bespoke billable engagements. Approved Organisations buying one can extend to another with documented incremental cost.
BritiAI is asking for a place on Lot 5 so that NHS Approved Organisations can run sovereign model training, agentic trial support and federated multi-Trust research with a UK supplier whose research credentials are on the academic record, not in the marketing deck.