iiinova — where your digital self keeps evolving

James Kermode @jrkermode.co.uk · 25/03/2026

MIT licensed, on PyPI (`pip install mograder`), docs and live demo: jameskermode.github.io/mograder/

jameskermode.github.io

James Kermode @jrkermode.co.uk · 25/03/2026

The slight irony: it was mostly written with Claude Code. Assessment tool for AI-assisted coding, built with AI assistance. Same patterns throughout: spec first, short verifiable tasks, close supervision. The overall design and assessment philosophy is the part that couldn't be delegated.

110

James Kermode @jrkermode.co.uk · 25/03/2026

Recntly added: hidden tests, late penalties, and progressive hints that students have to choose to open.

100

James Kermode @jrkermode.co.uk · 25/03/2026

The pipeline, insprired by Jupyter's nbgrader, but adapted for Marimo and tweaked pedagogic philospophy: • mograder generate: strip solutions, embed cell hashes • mograder autograde: sandboxed execution, integrity check against source • mograder feedback: HTML export with marks and comments

100

James Kermode @jrkermode.co.uk · 25/03/2026

mograder's answer: formative autograding on the mechanics (instant feedback), written component for understanding, backed up by randomly assigned oral mini vivas You can't delegate explaining code you don't understand.

100

James Kermode @jrkermode.co.uk · 25/03/2026

The problem it solves: students are writing code with AI assistance. That's fine, it's the new reality. But how do you assess it?

100

James Kermode @jrkermode.co.uk · 25/03/2026

One command. No Python install, no venv to setup, no config. The dashboard appears. And that's the most involved option. There's also a point-and-click desktop app, and a hub mode where students log in via SSO and edit in the browser with nothing local at all.

100

James Kermode @jrkermode.co.uk · 25/03/2026

I built a @marimo.io-native autograder for computational science courses. Here's what student setup looks like. 👇

230

James Kermode @jrkermode.co.uk · 22/01/2026

Fully-funded PhD at @hetsys.bsky.social: ML for Fracture in Fusion Materials 🔬 Develop #MLIP for brittle-ductile transition in W for fusion, with me, @pbro.bsky.social, Albert Bartok-Partay and @erikbitzek.bsky.social Deadline: 28th Jan warwick.ac.uk/fac/sci/hetsys/themes/projects2026/2026-008/

$Illustration of transition from low temperature brittle behaviour to high temperature ductile fracture in tungsten$

011

James Kermode @jrkermode.co.uk · 27/08/2025

Great to be at #Psik2025 in Lausanne along with a strong contingent from @hetsys.bsky.social. I am giving talk C1.09 at 17:45 today in Room 1 on fine tuning foundation MLIPs for applications in chemomechanics of structural materials- do come along if interested!

Students and staff from the HetSys CDT outside the Psi-k 2025 conference in Lausanne

070

James Kermode @jrkermode.co.uk · 08/07/2025

I enjoyed speaking at the QSP-UK Knowledge Exchange Workshop at Warwick this afternoon. I talked about uncertainty quantification, predictive modelling and scientific machine learning and their potential future application to #pharmacokinetics. #Marimo notebook at github.com/jameskermode...

lnkd.in

This link will take you to a page that’s not on LinkedIn

010

James Kermode @jrkermode.co.uk · 23/04/2025

For more on conformal prediction and its applications to quantifying uncertainty in quantities of interest computed with machine learning interatomic potentials see this paper by recently completed PhD student Iain Best in my group doi.org/10.1063/5.02...

doi.org

Uncertainty quantification in atomistic simulations of silicon using interatomic potentials

Atomistic simulations often rely on interatomic potentials to access greater time and length scales than those accessible to first-principles methods, such as d

020

James Kermode @jrkermode.co.uk · 23/04/2025

I am impressed with #marimo which lifts many of the limitations of #Jupyter notebooks: hidden state between cells, difficulty with version control, hard to make interactive UIs. The export to static WASM-backed HTML webpages is a very nice touch for sharing demos without losing interactivity.

110

James Kermode @jrkermode.co.uk · 23/04/2025

As part of the preparation of my lecture I tried out #marimo notebooks for the first time to make a demo of conformal prediction to address "collapsing posterior" problem of Bayesian inference in the high-data, low-noise limit. Published as an app here: kermodegroup.github.io/conformal-pr...

kermodegroup.github.io

conformal-prediction

a marimo app

110

James Kermode @jrkermode.co.uk · 23/04/2025

Visiting IESC in beautiful Cargèse, Corsica forTOTEMIC Training School. Great to give a lecture on multiscale materials modelling in age of machine learning to engaged audience of early career researchers, to meet new people, and catch up with friends and colleagues www.eera-set.eu/events/4969:...

eera-set.eu

TOTEMIC Training School 2025: "Tools for Energy Materials Modelling ...

Organised by the Cost Action EU-MACE project, the TOTEMIC “Tools for Energy Materials Modelling Acceleration” Training School 2025 is set to take place from 20 to 26 April in Corsica, France. TOTEMIC....

140

James Kermode @jrkermode.co.uk · 03/04/2025

That’s a wrap on our CECAM multiscale workshop. As organisers we were very happy with event, which has continued to nurture this vibrant and interdisciplinary community, nucleated in part by Sandro De Vita. Personally I was very proud of all the HetSys students for their excellent presentations!

010

James Kermode @jrkermode.co.uk · 31/03/2025

In the opening sessiont this afternoon, which is dedicated to our friend and mentor Sandro De Vita, we will be looking back on the progress made in the field since he sadly passed away in 2018

120

James Kermode @jrkermode.co.uk · 31/03/2025

We’re all setup and ready to go for this exciting CECAM workshop on getting information out of the atomistic scale (and the ideally doing something useful with it!), with a fantastic line up of speakers and participants www.cecam.org/workshop-det...

cecam.org

CECAM - Fulfilling the Multiscale Promise in Materials: Getting Information out of the Atomistic Scale

150

James Kermode @jrkermode.co.uk · 02/12/2024

We're now open for applications for 2025 PhD projects in the HetSys CDT at Warwick. Come and join us here!

050

Reposted by James Kermode

HetSys CDT at Uni of Warwick @hetsys.bsky.social · 02/12/2024

HetSys is shaping the next generation of experts to push the boundaries of computational modelling in a variety of complex, heterogeneous systems. We have a range of fully funded PhDs starting in October 2025. Explore the possibilities and get in touch #EPSRC #computationalmodelling bit.ly/3OA76tP

bit.ly

Available Projects for 2025 entry

062

James Kermode @jrkermode.co.uk · 01/12/2024

I’m having a go at Advent of Code in Rust this year to try to learn the language. Finished the day 1 puzzle at least 😀. I’ll post my solutions at github.com/jameskermode...

github.com

GitHub - jameskermode/AdventOfCode2024

Contribute to jameskermode/AdventOfCode2024 development by creating an account on GitHub.

020

James Kermode @jrkermode.co.uk · 26/11/2024

Thrilled that two PhD students from my group and #HetSysCDT passed their vivas with minor corrections: Lakshmi Shenoy today, and Iain Best two weeks ago. Thanks to @erikbitzek.bsky.social, Tom Hudson, @pbro.bsky.social and Michael Herbst for acting as examiners and congrats to Lakshmi and Iain!

Collection of prosseco corks signed by PhD students from my group after their vivas

041

James Kermode @jrkermode.co.uk · 24/11/2024

Like many others, I’ve made one of these world clouds from scholargoggler.com. Very happy with the result, represents the key themes of my research very nicely.

$Word cloud from scholar goggle for James Kermode: atomistic simulations, multiscale materials, hybrid fracture all listed.$

010

James Kermode @jrkermode.co.uk · 16/11/2024

Great to be here on BlueSky: feels like a breath of fresh air after the other place. I'm an academic interested in electronic/atomistic materials modelling. I've found many friends from computational materials science, #ai4science and #compchem active here, and looking forward to making many more!

0131