Skip to content
News
Event · October 14, 2026 · ~3 min read

Three days at Leeds, Manchester and York

Starting a collaboration with bioinformatics cores, on running Inflexa on their own infrastructure with their own models

Radu Tanasa at the lectern in a Manchester lecture hall, with the audience in the foreground and a slide asking what you would need to see before you trusted an agent's analysis enough to publish it.
The question we put to each room: what would you need to see?

In mid-October, Radu Tanasa spent three days at the University of Leeds, the University of Manchester and the University of York. The goal was to start working with their bioinformatics core facilities on one question: what would it take for a core to use Inflexa on its own infrastructure, on premises, with models it hosts itself?

This is the very beginning of that work. Nothing is deployed yet, and most of the answers are still ahead of us. We came back with good conversations, and with people we hope to work it through with.

  • University of Leeds

    Monday, October 12
    Talk and discussion at Omics Leeds

  • University of Manchester

    Tuesday, October 13
    Research seminar, then meetings

  • University of York

    Wednesday, October 14
    Meetings on working together

Leeds: what would you need to see?

On Monday, Radu spoke to the Omics Leeds community at the invitation of Elton Vasconcelos. The talk was about AI agents in computational biology, and what it would take to actually trust an analysis one of them ran for you. About half the hour went to discussion, around one question: what would you need to see before you'd publish an analysis an agent ran?

Most answers were along these lines:

  • “I’d want to be able to re-run it independently.”
  • “I’d like to be able to see the reasoning behind each decision.”
  • “I’d want to make sure reports and references are not hallucinated.”

All three come back to the same thing: being able to check the work yourself. It is why the scientist has to stay in charge. An agent can take on more of the running, and a person still makes the calls and puts their name on the result.

Radu Tanasa in front of the floor directory of the Worsley Building at the University of Leeds.
At the Worsley Building, Leeds.
Radu Tanasa speaks to a seated audience, next to a slide titled Then someone asks three questions.
Three questions: why this method, where did this number come from, can someone else rerun it?
Radu Tanasa at a laptop while another man holding a coffee cup talks through a point beside him.
Talking it through at the laptop.

Manchester: from the room to the next step.

On Tuesday, Radu presented at a University of Manchester research seminar at the invitation of Leo Zeef. It turned into the same conversation as in Leeds: people want to use AI in their analysis, and they want work they can check and stand behind.

The seminar was only the start of the day. The rest of it went to meetings about how to make Inflexa available at Manchester. Those were first conversations, and they are the step we care about most: going from a good discussion in a room to something people can actually use in their work.

Radu Tanasa with a coffee outside a University of Manchester building with a molecular mural behind him.
Arriving at the University of Manchester.
Radu Tanasa at the lectern, looking up at the opening slide, From AI Chat to Reproducible Computational Biology Workflows.
From AI chat to reproducible computational biology workflows.
View from the back of a lecture hall, with rows of attendees watching a slide titled A chat transcript records what was typed.
The view from the back of the hall.
A man with a microphone speaks from the front of the hall while Radu Tanasa listens at the lectern, under a closing slide reading Fast is easy now. Checkable is the work.
The closing slide: fast is easy now, checkable is the work.

York: the same answer.

On Wednesday, Radu met Alastair Droop, Sandy Macdonald and Matthew Care at the University of York.

In all three places, we ended up talking about how we could work together, and each time the conversation converged on the same answer: start with the core.

Why the cores.

A lot of the analysis in a university runs through a small team of experts in a core facility. The data keeps growing, and now there is an expectation that AI should make everything instant. Every core we have spoken to is dealing with some version of that.

Fast is easy now. Trustworthy is still the hard part.

We would love to see core teams spend less time on routine runs and more on the judgment only they can bring, with every result easy to trace back and explain. A core's data, its compute and its ways of working are already in place, so whatever we build has to fit around them. That is why on-premises infrastructure and on-premises models are where this starts.

What comes next.

The only way to build this is with the people who run these cores, around how they really work. The next step is to keep those conversations going, and we will share more when there is more to share.

Thank you to Elton, Leo, Alastair, Sandy, Matthew and everyone who came along, for your time and your hospitality.

Run a core, or a group with data waiting for a paper?

We'd like to keep the conversation going, even if it's just to hear what's missing.

Get in touch