What if the hardest part of AI in pharma isn't the AI?
I've spent months trying to answer one question: how do we actually build Pharma 5.0?
Read on LinkedIn →
Seven Principles for Pharma 5.0 · GxP Validation · Local AI
GxP-ready AI, with the evidence to prove it
I build AI systems for GxP use with their complete validation package, and I help pharma and biotech teams validate the AI systems they already have. Everything runs on your own infrastructure, and your quality unit stays in authority.
Author of the Seven Principles for Pharma 5.0: an independent framework for human and machine collaboration in regulated drug development, aligned with the FDA and EMA guiding principles for AI. Read the principles →
Working with leading pharma and biotech
Seven Principles for Pharma 5.0
Multiply what experts can do, accountable through evidence
The FDA and EMA ten principles describe how to make AI safe across the drug product life cycle. My seven build on them to make Pharma 5.0 happen: each one a single commitment, with the problem it solves and the practices that make it concrete. Drawn from my research and from working with pharma and biotech teams on these problems.
IPurpose
AI exists to lift what experts can do, and we measure that lift against how the work is done today.
IIPartnership
Humans and machines each do what the other cannot, and answer to each other for it.
IIIIntegrity
Build the machine so its core cannot be corrupted, and so its work can always be questioned.
IVEvidence
Prove that the human and the machine together get it right, and keep proving it.
VPeople
Grow the people who can both use the machine and judge it.
VICapability
Let those who know the work build for it, on a platform that keeps it defensible.
VIIResilience
Stay in control of everything the partnership depends on.
Read the Seven Principles →
Version 1.1, October 2026. An independent proposal, not endorsed by any regulator.
Latest from LinkedIn
Notes from the work
In the age of AI, software is not the commodity.
Humans are. Sit with that for a second, because it sounds backwards. For decades, software was the scarce thing. It took teams, years, and serious budgets to build. That's ending.
Read on LinkedIn →It's Thursday, and here's another lab spreadsheet we turned into a free, open-source tool.
One more step towards Pharma 5.0, which to me means software that does the checkable work and shows it, so the scientist stays in charge. So what does this one do?
Read on LinkedIn →The Pharma 5.0 Thesis
From digitization to collaboration
Pharma 4.0 was digitization. It moved paper to systems, batch records to electronic logs, manual review to dashboards. Necessary work, and still unfinished. The records are digital, yet many of the calculations and checks that decide things still happen in spreadsheets and by hand, where a number can be wrong in ways that look right.
Pharma 5.0 is human and machine collaboration, with accountability on both sides. It can only be built on work that is already digital and defensible. So the path has two steps: finish Pharma 4.0 with open tools anyone can use, then bring AI into the workflow on your own infrastructure, with experts making every consequential decision.
Speed.
Compress weeks of expert work into hours, without losing the expert. Cycle time falls because retrieval
and first drafting happen in minutes, not days. Experts are redirected to the consequential
decisions, where their judgment has always been the limiting factor.
Accuracy.
Match or exceed consulting-grade precision, measured against expert-annotated benchmarks. Citation
precision, hallucination rate, source coverage, and confidence calibration are tracked as explicit
metrics, not afterthoughts.
Accountability.
Every claim traces to a timestamped source. Every agent decision is reviewable. Every human override is
logged. In a regulated environment, the answer must be defensible under audit six months or six years
after the fact. Provenance is the foundation, not a feature added at the end.
Free Tools
Free, open-source tools for Pharma 5.0
Access to Pharma 5.0 should not depend on budget. I am committed to building a series of free, open-source tools that finish the work of Pharma 4.0: deterministic, transparent, and running in your browser or on your own computer, with nothing you type leaving your machine.
01
Ligant bench tools
Four browser tools for cell and gene therapy labs: an Antigen Density Calculator, a Molarity Converter, an Antibody Titration Planner and a Dilution Planner. Each states its assumptions and refuses to return a number the inputs do not support.
Free · Apache 2.0 · Runs in your browser
02
ABM Invoice Analytics
A desktop app that checks every invoice against everything billed before and flags the duplicates. It runs entirely on your computer, and every flag shows exactly why it was raised.
Free · Windows, Mac, Linux · Source code public
03
ABM Local AI
A private AI server for your own computer. It downloads, verifies and runs models so ABM’s free apps can use them, and it listens only on your machine.
In development · Apache 2.0
Local AI
Private AI, trained on your data, on your premises
For work that cannot leave the building, I help pharma and biotech companies build and deploy their own AI, trained on proprietary data and running on their own infrastructure. Every use case starts with a declared context of use and a declared status: critical or non-critical.
01
The right model for the risk
Frozen, deterministic models for critical use, with determinism measured on the deployed environment rather than assumed. Small language models only for non-critical work such as summaries, under continuous human oversight, and never on the critical path. Numbers come from the deterministic tool, not the model.
Critical or non-critical, declared for every use case.
02
Your data, your model, your premises
Models, reference data and workflows installed on your own servers and workstations. That data stays inside your environment is verified at the network level on the deployed hardware, not inferred from a configuration file. A model trained on your data stays in your environment and never informs another engagement.
Built with your IT and security teams.
03
Validation evidence for regulated use
I produce the validation evidence against Annex 11 and your own CSV/CSA framework, designed for readiness under the draft Annex 22 on artificial intelligence, with test data kept independent of training data. Your quality unit executes and approves it.
For QA and validation departments.
04
Coaching for IT and QA
Working alongside IT and quality teams on what validation means for an AI system, and on where to narrow what a model is permitted to decide.
Often the most valuable part of the work.
Or start with the Seven Principles for Pharma 5.0, or the ten FDA and EMA principles they build on.
How I Build
From requirement to local app
01
Your requirement
Bring the workflow you want to improve, the data it touches and the constraints that matter. Together we declare its context of use and whether it is critical or non-critical.
02
URS, drafted and approved
My AI agents, running on local AI, draft the User Requirements Specification. Your subject-matter expert reviews and approves it. Nothing is built until they do.
03
Built and delivered
I build the local app to the approved URS and deliver it onto the GPU hardware we set up for you, inside your own environment.
04
Deployed for your team
The app goes live locally for everyone who needs it, with validation evidence for your quality unit to execute and approve, and change control for every update.
Let’s Talk
Bring me your questions
Whether you’re curious about Pharma 5.0, working out how to validate an AI system in a regulated environment, or wondering what local AI can already do for pharma and biotech teams today, I’d like to hear from you.
Pharma 5.0 · Validating AI in GxP · Local AI for pharma and biotech
You don’t need a project in mind. Send a question, an idea, or the problem you’re working through, and I’ll reply personally. The more people who understand what’s possible, the faster this industry moves.