Dr Oliver Fisher
“It won’t write it for you. It gives you feedback to takeaway and think about.” Dr Oliver Fisher, EDI+ Fellow, the University of Nottingham
Picture the scene. The deadline is getting close. The research plan is written, the budget has been worked through, and the collaborators are in place. Most of the application is done. Then you reach the EDI section, and pause.
You want to write something useful, something that belongs to the project rather than sounds bolted on and generic. But the guidance is spread across funder pages, university systems and old training notes, and as the deadline approaches, it can be hard to know what a good answer should look like.
That was the problem EDI+ Fellow Dr Oliver Fisher, Assistant Professor in Chemical and Environmental Engineering at the University of Nottingham, kept hearing from colleagues.
“It came out of talking to colleagues who quite often knew EDI was an important topic, but didn’t feel they had the expertise to do something meaningful with it,” he recalls. “And a bit of frustration that when they went to search for that information, it was all fragmented.”
Through his EDI+ Fellowship, Dr Fisher has been turning that frustration into a practical tool: the EDI Action Plan Coach. A researcher enters a brief project summary and a draft EDI plan, then selects their funder and institution. The tool provides structured feedback: questions to reflect on, gaps to consider, relevant funder expectations and discipline-specific prompts. What it gives back is not a finished paragraph, but a set of questions and challenges for the researcher to work through.
A tool that asks, not answers
“It won’t write it for you,” Dr Fisher says. “It gives you feedback to takeaway and think about.”
It’s a deliberate choice that comes with AI’s well-publicised downside, bias. Large Language Models that underpin generative AI, such as ChatGPT or Claude, can reproduce assumptions based on the information on which they are trained. Put simply, the outputs reflect the input. Based on Google’s Gemini, Dr Fisher does not pretend that the Coach is immune from the same biases, nor that they will magically disappear because it is being used for EDI.
“That is definitely a limitation of the tool,” he says, “and that’s why the human-in-the-loop element is so important: people still need to evaluate what comes out of it critically.”
The researcher remains responsible for the judgement. The Coach can suggest, prompt and challenge, but it cannot decide what inclusion should mean in a particular project. Dr Fisher has also been looking at bias-detection tools as another layer of checking, but the final checkpoint remains human.
The technical risk matters. So does the very human one: that someone under deadline pressure uses the tool to finish the box rather than think it through.
“That’s my biggest fear,” he says.
Still, Dr Fisher is quick to point out that researchers who want a shortcut can already get one from a general AI tool. Nevertheless, he hopes the Coach does something slower and more useful: gives them enough of a pause to think again.
“This is hopefully getting them to reflect: ‘I hadn’t thought about that before; I could do this,’” he says.
A tool that changed shape
The tool changed as Dr Fisher built it. His first instinct was to train a bespoke model on real examples: draft EDI sections paired with stronger final versions, so it could learn what good looked like. But that approach would have required hundreds of real proposals, and the data were not available, certainly not within a part-time fellowship. If even the person building the tool could not easily find good examples, it was not hard to see why researchers might struggle with that part of the form.
Changing tack, Dr Fisher decided to develop the tool around Google’s Gemini, using structured prompts and a set of reference points to keep it anchored: funder criteria, individual funders’ EDI priorities, and resources from networks such as the EDI Caucus. When someone uses it, their draft is checked against that material before the tool returns feedback.
He also needed a way to test the quality of that feedback without asking 50 or 100 people to review it each time. An MSc student’s project offered a solution by adopting an emerging evaluation approach increasingly used in AI research: give several language models the same task, then ask them to review one another’s answers against funders’ criteria. It was not a perfect substitute for human judgement, but it gave Dr Fisher a quick way to see where the feedback was stronger, weaker or drifting off course.
A researcher changed by the work
The Fellowship has also fed back into Dr Fisher’s main research. His day-to-day work uses machine learning in engineering, including developing low-cost sensors for wastewater and process manufacturing. Part of the appeal of that work is access: finding ways to put expensive technology within reach of smaller organisations that might otherwise be priced out.
Working on EDI has made him look at that ambition more closely. A sensor may be cheaper, but is the interface usable? Are the colours accessible for colour-blind users? Can different people work with the system easily? Are the students and researchers building it getting proper development, recognition and ownership of their work?
“They’re all questions the EDI+ Fellowship has made me think about,” he says
That may be one of the unexpected outcomes of the Fellowship. It has not only produced a tool for researchers writing funding proposals, but also a tool for researchers writing funding proposals. It has changed the questions Dr Fisher brings back to his own engineering work.
A place to begin
A second version of the Coach is due by the end of the project, with an open-access release planned based on feedback from human users. Dr Fisher has also begun conversations with EPSRC about what the tool might become. With further funding, he would like to build in more continuous human-in-the-loop learning, closer to the system he first imagined.
But for now, his measure of success is simpler. It is the researcher staring at the EDI section of a proposal, unsure where to begin.
“I hope this is a tool that’s going to make writing those sections less intimidating,” Dr Fisher says, “especially for those early in their career, who are aware of the importance of EDI and want to do something about it but don’t know where to start. It can ask better questions and offer a place to begin, but it can’t do the thinking for them. That’s the point.”