About T2R2
Transparent, Traceable, Replicable, Reproducible.
T2R2 is a platform for regulated software in pharma and healthcare. It builds the application, it runs inside the application, and it keeps the application's AI agents under control. We have been building it since 2022.
1 · The platform
It builds your application.
One linked chain runs from the intended purpose to the release. AI agents draft the requirements, risks, specifications, code and tests. Named people approve at every gate. Each release carries its Evidence Pack.
2 · The library
It runs inside your application.
The T2R2 library powers every application built on the platform. It keeps the record, checks every permission, and holds the point where a person decides. Each application starts with these foundations instead of writing them again.
3 · The agentic AI harness
It keeps AI agents under control.
A harness is the structure that holds AI agents in place. It decides what they may touch and records what they do. Any model can run inside it, including one installed on your own servers.
The agentic AI harness
The frame around the AI.
Hospitals and pharmaceutical companies are not short of AI models. What they need are answers to three questions on the day something goes wrong. Did the software behave the same way yesterday? Can you show, years later, what it saw, said and did? Where did the data go?
T2R2 does not compete with the models. It is the frame that makes their use acceptable in a regulated setting.
AI may propose. Only people decide.
Nothing that changes a patient's care happens without a decision by a named person. This is built into the structure of the software, not added as a confirmation button.
Everything is on the record.
Every observation, proposal, decision and answer is written to a journal in which each entry is chained to the one before. A whole episode can be replayed exactly as it happened.
Data goes only where it is cleared to go.
Each model is cleared for a class of data. Identified patient data cannot reach a model that is not cleared for it.
No silent upgrades.
Every agent is pinned to a model version and a prompt version. A new version ships only after it passes its evaluations, and the change is recorded.
Explicit failure.
When a model fails or its budget runs out, the system says so on the record. It never uses a partial answer and never switches model silently.
The library
The record underneath every application.
Every application built on T2R2 inherits the same core. The core is deliberately small, so that the part which carries the heaviest scrutiny can be read and reviewed.
Every change is an event.
The state of the application is always computed from its journal, the way a balance is computed from the entries in a ledger. Any past state can be shown as it was.
The record is tamper-evident.
Entries are chained. An altered entry breaks the chain, and the system refuses to start on a broken chain.
Erasure keeps the audit trail whole.
Personal data is encrypted with its own key. Destroying the key erases the data, and the record of what happened stays intact.
Every action needs a permission.
No component and no agent has standing authority. Each use of a permission is checked and recorded.
It runs where you decide.
T2R2 runs on standard infrastructure, in your own data center first and in the cloud when you choose.
Designed for the most demanding case
One design, for the strictest profile an application could bring.
T2R2 is designed for medical-device software at the highest class and for high-risk AI. An application with a lighter profile fits underneath. Certification and validation belong to each application built on T2R2, and to the organisation that owns it.
Medical devices and AI, evidence for: EU MDR · IEC 62304 · ISO 14971 · IEC 62366-1 · IEC 81001-5-1 · EU AI Act · ISO 13485.
GxP, built for the expectations of: GAMP 5 · 21 CFR Part 11 · EU Annex 11. Data protection: GDPR.
History
Built since 2022.
T2R2 began with one requirement from pharmaceutical research: a result that cannot be reproduced is worth nothing. Everything since then has grown from it.
- 2022
T2R2 starts as a distributed multi-agent system. Each agent runs one step of an AI or data-science pipeline and records what it did, so the whole pipeline is transparent, traceable, replicable and reproducible by design. The name dates from that year.
- End of 2022
T2R2 goes into use for pharmaceutical companies and contract research organisations, where it runs data and AI pipelines.
- 2024
We begin the second generation. The same principles are rebuilt for regulated software, for patient data, and for AI agents built on large language models that must answer to people.
- 2026
The second-generation core is in place. The platform builds its first regulated applications, in medical-device software and in clinical research.
What it enables
One foundation, many applications in pharma and healthcare.
- GxP applications that no product serves, at manufacturers and CDMOs.
- Patient applications for clinical trials, such as electronic consent, diaries and long-term follow-up.
- Clinical applications that hospitals bring into routine care as their own medical devices.
- Software as a medical device, built with its technical file.
- AI agents inside any of these, each one recorded, replayable and answerable to a person.
T2R2 also enables the Patient Digital Twin, a companion for patients and care teams developed by LLAAMA. Read about it on llaama.com →
Questions
Frequently asked questions
What is T2R2?
T2R2 is a platform for regulated software in pharma and healthcare. It builds GxP and medical-device applications fitted to your process, it runs inside them as a library, and it keeps their AI agents under control. The name stands for Transparent, Traceable, Replicable and Reproducible.
How does T2R2 make regulated software faster?
Software and compliance used to be two projects, run side by side by two teams and joined by hand at the end. T2R2 makes them one piece of work: every requirement, risk, specification, test and release belongs to one linked record, written as the software is built. The work is done once, not twice.
Is an application built with T2R2 validated or certified?
Validation and certification belong to each application and to the organisation that owns it. T2R2 delivers the evidence with every release, in an Evidence Pack, and your named experts review, approve and sign it.
What is in an Evidence Pack?
For a GxP application: user requirements, risk assessment, specifications, trace matrix, test evidence and release manifest. For medical-device software: intended purpose, requirements, risk file, threat models, architecture, trace matrix, test evidence and release manifest. The documents use the T2R2 templates and are mapped to your SOPs.
Which standards does T2R2 support?
For medical devices and AI, T2R2 produces evidence for EU MDR, IEC 62304, ISO 14971, IEC 62366-1, IEC 81001-5-1, the EU AI Act and ISO 13485. GxP applications are built for the expectations of GAMP 5, 21 CFR Part 11 and EU Annex 11.
What is an agentic AI harness?
A harness is the structure that holds AI agents in place. In T2R2 it decides what each agent may touch, requires a human decision before anything changes a patient's care, pins every agent to a model version and a prompt version, and records every exchange with a model so that it can be replayed exactly.
Which AI models can run in T2R2?
You choose the models and where they run, including a model installed on your own servers. Which model may see which class of data is a deployment setting, and that setting is itself recorded.
Where does an application built with T2R2 run?
On standard infrastructure, in your own data center first, and in the cloud when you choose.
Who owns the application?
You do: the application, its documentation and its data. We are preparing the T2R2 core for open-source release under the Apache 2.0 licence.
How long has T2R2 existed?
T2R2 started in 2022 as a distributed multi-agent system that ran AI and data-science pipelines, traceable and reproducible by design. It has been used by pharmaceutical companies and contract research organisations since the end of 2022. The second generation began in 2024.
Does T2R2 replace our existing systems?
No. T2R2 builds the applications your process needs where your existing systems do not fit, and connects them to the systems you keep, such as laboratory results and the electronic health record. You own the result, and we keep it compliant with you as it changes.
How do we start?
With an early-adopter pilot: one application, one fixed-length cycle, a fixed price and exit criteria agreed at the start. At the end you keep the application with its Evidence Pack, reviewed by your own team.