When Jenny Blakesley heard about the AI for Ops programme, it prompted a new way of thinking about a long-standing operational challenge. As one of the Deputy Heads of Cambridge University’s Education Services, she was aware of the increasing complexity facing colleagues in the Exams team each year.
At the centre of this challenge sits a detailed and intricate set of Statutes and Ordinances—in parts centuries-old, leather-bound volumes that form the University’s constitutional framework. These regulations determine the combinations of exam papers students can take to qualify for their University of Cambridge degrees.
The distinctive structure of the Cambridge undergraduate “Tripos” system offers students considerable flexibility. Unlike the modular systems used by most universities, Cambridge undergraduates across all disciplines—from classics to chemistry—can “borrow” papers from other subject areas to construct their Bachelor of Arts degree. However, only specific combinations of papers are permitted under the Statutes, and these combinations can change each year.
While this flexibility is a defining feature of the Cambridge experience, it also creates a significant administrative task. Each October, around 65,000 exam sittings must be checked, with every student’s choices reviewed to ensure they comply with the regulations. This process is time‑intensive and requires careful attention to detail. Crucially, until all exam standings are confirmed, decisions about scheduling and logistics cannot proceed.
“When I heard about this problem, I couldn’t ignore it,” says Jenny. “I felt there must be a better way of doing it.”
From manual checks to machine learning
A potential solution emerged through the AI for Ops funding call, part of the ai@cam initiative supporting professional services teams to apply AI in improving University operations. Jenny recognised the opportunity to rethink how this process could be done.
Her proposal was straightforward but ambitious: to create a machine-readable version of the Statutes and Ordinances that could replace the largely manual exam-standing process. By using large language models (LLMs), such as Copilot, ChatGPT or Gemini, she identified a way to translate complex text-based regulations into a reusable “rules library”. This could reduce manual checking time significantly, with AI verifying exam combinations quickly and allowing the team to focus on resolving exceptions.
“It will likely take less than a minute to produce a report that confirms whether exam combinations are valid,” says Jenny. “That could save weeks of work and enable faster responses to students, as well as more efficient timetable planning. Ultimately, it improves the overall student experience.”
When Jenny was matched with a machine learning mentor through the programme, she quickly saw the practical potential of an AI-driven approach. A key priority, however, was to ensure the solution would be transparent, maintainable, and not dependent on expensive external support.
Using AI for Ops funding, she recruited a recent graduate to develop the system in Python. Alongside the code, a step-by-step “Haynes manual” is being produced, enabling colleagues with a working knowledge of the exam-standing process to update and maintain the system as regulations evolve. This approach supports long-term sustainability.
At the same time, the team has been testing a range of LLMs, benchmarking their outputs against previously produced manual reports to assess accuracy and reliability. “One of the strengths of this project is that we can test it rigorously,” says Jenny. “We have years of historical data, where exam combinations have already been checked manually. That gives us a strong basis for validating the system.”
If successful, the ambition is to extend the solution across all 33 undergraduate degree programmes, with the aim of having the AI-supported process in place as soon as practicable—ideally for the next academic year. In the longer term, similar approaches could be applied to postgraduate exams and other areas of the University that rely on complex regulatory frameworks.
“What’s been particularly valuable about the AI for Ops programme is the opportunity to reduce some of the more routine aspects of the work,” says Jenny. “That creates space to focus on the more complex cases and for strategic questions around education and student experience.
“Being able to synthesise information at this scale opens up a range of possibilities. The programme has provided both the support and the freedom to explore what’s achievable.”