Operator-Founder · Building in the Open
For two years I put AI to work inside other people's companies.
Here is what it looks like pointed at my own.
In workshops and boardrooms I have said the same thing, again and again: AI belongs in the way a company is run, not in a pilot project.
At some point the question became unavoidable: would I do it myself?
The answer is Pheraxia GmbH in Freiburg: formulation modelling, one molecule at a time.
The landscape
The tools exist
Poor aqueous solubility has been worked on for decades, and well. Amorphous solid dispersions, lipid-based and self-emulsifying systems, nanocrystals, cyclodextrins, salt and co-crystal forms are established and in industrial use. We are not offering a replacement for them. We do not work on biologics and we do not work on minerals.
The established routes are answers. Which step still holds a particular molecule back is a different question.
What we do
One molecule. One question.
Pheraxia · Greek φέρω (to carry) + ἀξία (worth, excellence) · “the worth and excellence of carrying”
For ingredient, formulation and contract manufacturing businesses: we find which step holds a molecule back and design the experiment that would test it.
A formulation that fails does not fail everywhere. It fails at a particular step, and finding that step is the work.
We do not replace your formulation team; we start where its last attempt ended. We work on data only — no physical material ever comes to us.
A molecule already characterised in your own data, now stalled at the formulation stage.
Routes already tried are not lost work: each one narrows where the limiting step can be.
Mechanistic: it works from the molecule in front of us.
We design with established, well-documented excipients rather than novel materials.
We work on data: what your team already has, and what the test produces.
Can the target performance be reached within the stated assumptions? If not, we say so.
How it works
Three stages. One report.
- The question
We model the limiting step against the performance your existing formulations already reach.
- The experiment
Before anything runs, we write down the assumptions, the design and the question the test must answer.
- The reading
The test runs at an agreed laboratory; your material goes there directly. You receive one written report.
One written report. You can act on it even if the answer is no.
Why this belongs to my advisory work
For most of those two years I watched the same pattern: AI as a side project. A pilot. A buzzword. Never the backbone.
At Pheraxia I work the way I recommend to my clients: AI in the daily workflow; decisions stay with people. There is a co-inventor on the science side. People decide, not the model.
This is not about automation. It is about scaling human judgement.
That is exactly the substance of my AI empowerment work — except here I do not recommend it, I carry it. What I learn flows straight back into the work with my clients, and into the blogand video on this site.
Tell us about the problem, not the molecule.
A first conversation needs no structures and no formulations — only the problem you are trying to solve.