Oxbridge MentorsSPARK
Chat on WhatsApp

Oxbridge Mentors × SPARK

The AI Research Fellowship

An eight-week, fully remote research project run by SPARK for Oxbridge Mentors students. You join a team of two to four with a PhD mentor from institutions including Cambridge, Imperial, Google DeepMind, Meta, and the AWS AI lab. You will work on a real AI research question, write code, run experiments, and submit to a top-tier international AI conference. It runs part-time, at around 8–10 hours a week.

Register your interestSee the timetable
Price
£2,000
Length
8 weeks
Commitment
8–10 hrs/week
Next start
2 Nov 2026
Format
Fully remote

What you walk away with

Eight weeks of hands-on, publishable AI research

Each week has a lecture, a mentoring session on your project, and a hands-on workshop. By week eight your team has a paper.

A research paper with your name on it

Your team takes one question from first reading to a full write-up, prepared for submission to a top AI conference.

Real technical skill

You write and run your own code, call and fine-tune large language models, design experiments, and analyse the results.

A mentor who knows the field

Weekly sessions with a PhD researcher who publishes at the venues you are aiming for, plus Slack access between sessions.

Something substantial to talk about

A project you understand deeply and can discuss in detail, in an interview or an application.

Dates & pricing

What it costs and when it starts

£2,000 covers the full eight weeks. The next cohort starts Monday 2 November 2026.

Oxbridge Mentors rate

£2,000

per student, for the full eight weeks

  • 8 weekly lectures on core AI concepts
  • 8 weekly team mentoring sessions with your PhD mentor
  • 8 weekly hands-on research workshops
  • Slack access to the mentoring team between sessions
  • A research paper prepared for conference submission
  • Certificate of completion

Scholarship support is available. Message us on WhatsApp to discuss access support or payment options.

Register your interestAsk about dates

Late autumn 2026

Next intake

Starts Monday 2 November 2026

Register interest by

Monday 26 October 2026

Programme ends

Friday 25 December 2026

Spring 2027

Starts Monday 4 January 2027

Register interest by

Monday 28 December 2026

Programme ends

Friday 26 February 2027

Late spring 2027

Starts Monday 5 April 2027

Register interest by

Monday 29 March 2027

Programme ends

Friday 28 May 2027

Summer 2027

Starts Monday 5 July 2027

Register interest by

Monday 28 June 2027

Programme ends

Friday 27 August 2027

Each cohort runs the same eight-week structure. Teams are capped at four students, so places are limited. There is no application form: message us on WhatsApp and we will hold you a place.

The full eight weeks

Your week-by-week timetable

There are 3 sessions each week. Here is what they cover.

How each week runs

  • Lecture. A short foundational lecture teaching a specific concept (e.g. transformers, the full LLM training pipeline, world models).
  • Mentoring. A team session with your mentor focused on the research project you are working on.
  • Workshop. A hands-on session teaching a practical research skill you need that week.
  • Around 8–10 hours a week in total, covering the sessions and your own reading, coding, and experiments with your team.
  • Exact session times are flexible and agreed each week on Slack.

Further support

  • Async access. Your mentor is reachable on Slack between sessions.
  • Extra calls. Arranged around your availability when you need them.
  • All access to the mentoring team runs through Slack.

Weekly schedule

What each week covers

Mentoring sessions are project-specific. The focus shown for each week is typical; your mentor adapts each session to your team’s project and progress.

Week 1

Lecture· the concept

What is machine learning? From rules to neural networks. The big picture of how systems learn from data rather than being explicitly programmed: supervised and unsupervised learning, training versus inference, classification and generation, and why neural networks became so powerful for modern AI.

Mentoring· your project

Project framing and research direction. Meet your mentor and team, discuss the project direction, clarify the broad research question, agree ways of working, and divide up the initial reading. By the end of the week your team should have a shared Overleaf project, an initial list of relevant papers, and the start of a literature review.

Workshop· a practical skill

Research foundations:

  • Working effectively as a research team
  • Finding relevant papers (Scholar, arXiv, Semantic Scholar)
  • How to read an academic paper, and run a literature review
  • Citations & references in Overleaf / BibTeX
Week 2

Lecture· the concept

How ChatGPT works: transformers, tokens, and language models. The core ideas behind modern language models: tokenisation, embeddings, attention, transformer blocks, pretraining, and instruction tuning, and why large language models can generate fluent, useful text.

Mentoring· your project

From literature to a concrete plan. Discuss what the team has learned from the literature, identify the research gap, narrow the project scope, and agree an initial experimental direction. By the end of the week your team should have a clearly written research question, an initial hypothesis, and a first experiment plan.

Workshop· a practical skill

Getting set up for inference:

  • Setting up your Vertex AI account
  • Running API inference
  • Mapping your first set of experiments
Week 3

Lecture· the concept

Adapting and steering LLMs: prompting, fine-tuning, LoRA, RAG, and control. How general-purpose models are adapted for specific tasks and behaviours: prompting, few-shot learning, retrieval-augmented generation, supervised fine-tuning, parameter-efficient methods such as LoRA, and techniques for steering model behaviour.

Mentoring· your project

Building the first experimental pipeline. Review early implementation progress, debug the setup, decide what outputs to log, and agree how experiments should be recorded reproducibly. By the end of the week your team should have a first end-to-end prototype or experiment running, even if the results are still rough.

Workshop· a practical skill

Prompt engineering for API inference:

  • Few-shot prompting
  • Structured / JSON output and defining a schema
  • Tips for a reliable, reproducible code pipeline
Week 4

Lecture· the concept

Beyond text: multimodal AI systems. AI systems that work across text, images, audio, and video: vision-language models, image generation, speech systems, video understanding, and how multimodal systems let AI interact with the world in richer ways than text alone.

Mentoring· your project

Refining the experimental design. Review first results, identify problems or failure cases, improve the experimental setup, and plan stronger comparisons, baselines, metrics, or datasets as appropriate. By the end of the week your team should have a working pipeline and a revised plan for the main experiments.

Workshop· a practical skill

Evaluation, metrics, and error analysis:

  • Defining what counts as success or failure
  • Designing an evaluation rubric
  • Human evaluation vs LLM-as-a-judge
Week 5

Lecture· the concept

AI agents: tools, planning, and acting on the web. Systems that use tools, call functions, browse information, write code, and take actions for users: tool use, planning, memory, function calling, web agents, and the important question of when an agent should choose not to act.

Mentoring· your project

Generating core results. Run larger experiments, compare conditions, add baselines, carry out ablations, and begin error analysis where relevant. By the end of the week your team should have a first complete results table and initial observations about what the results show.

Workshop· a practical skill

Results and model failure analysis:

  • Review first experimental results
  • Identify where and why models fail
  • Categorise failure modes, e.g. hallucinated tool use, failure to abstain, poor handling of code-switched queries
Week 6

Lecture· the concept

Can we trust AI? Evaluation, uncertainty, and hallucinations. How we evaluate AI systems and decide whether their outputs can be trusted: benchmarks, accuracy, error analysis, calibration, uncertainty, hallucinations, abstention, and why evaluation is often harder than it first appears.

Mentoring· your project

Turning results into research claims. Discuss what claims are supported by the evidence, what extra experiments are needed, and how to structure the paper around the main contribution. By the end of the week your team should have a first full paper draft underway, including the method, experimental setup, and preliminary results.

Workshop· a practical skill

Writing up research: structure, figures, and tables:

  • How an AI conference paper is structured, section by section
  • Turning a results table into a claim a reader can check
  • Designing figures and plots that carry the argument
  • Writing a method section someone else could reproduce
Week 7

Lecture· the concept

When AI goes wrong: bias, safety, and adversarial attacks. Failure modes in AI systems: bias and fairness, misuse, jailbreaks, prompt injection, adversarial examples, robustness, and why safety is not just an ethical issue but also a technical research problem.

Mentoring· your project

Final experiments and paper revision. Resolve remaining technical blockers, complete any final experiments, strengthen ablations or error analysis where useful, and revise the paper for clarity, evidence, and narrative. By the end of the week your team should have completed the main experiments and produced a full draft ready for detailed review.

Workshop· a practical skill

Making results hold up: reproducibility and stronger evidence:

  • Seeds, variance, and repeated runs: what a single number hides
  • Choosing baselines and ablations that isolate your contribution
  • Sanity checks and common ways experiments quietly mislead
  • Organising code and configs so results can be regenerated
Week 8

Lecture· the concept

Can AI do science? World models, discovery, and the future of AI research. A look ahead: world models, reasoning, AI for coding, theorem proving, scientific discovery, automated research assistants, and whether future AI could help generate new scientific ideas.

Mentoring· your project

Polish, presentation, and submission readiness. Refine the argument, tighten figures and tables, prepare the final presentation, and make the paper as submission-ready as possible. By the end of the week your team should have finalised the paper and presented the project to the cohort.

Workshop· a practical skill

Submitting and presenting your research:

  • How AI conference submission and peer review actually work
  • Writing a title and abstract that land the contribution
  • Preparing the final presentation to the cohort
  • Reading reviewer feedback and planning a response

Session times are agreed each week and coordinated on Slack. Questions before you start? Message us on WhatsApp.

Who you work with

Your mentors

PhD researchers who publish at the conferences your team is aiming for, from the University of Cambridge, Imperial College London, Google DeepMind, Amazon, Meta, and the AWS AI lab.

Ivaxi Sheth

Ivaxi Sheth

CISPA · Imperial College London · Google DeepMind · Amazon

Works on causal reasoning in large language models and on trustworthy machine learning: how models handle cause and effect, and how fairness, robustness, and explainability trade off against one another. PhD at CISPA, previously Imperial College London, with research at Google DeepMind and Amazon.

Google ScholarLinkedIn
Vatsal Raina

Vatsal Raina

University of Cambridge · Meta

Works on question answering and retrieval: how models read a text and answer questions on it, how exam-style questions can be generated and marked automatically, and how to build efficient retrieval-augmented embeddings. PhD in the Machine Intelligence Laboratory at Cambridge, with industry research at Meta.

Google ScholarLinkedIn
Vyas Raina

Vyas Raina

University of Cambridge · AWS AI lab

Works on the robustness of large language models: the adversarial attacks that break them, and whether an LLM can be trusted to grade other models’ work. Published at ACL and EMNLP. PhD in the Machine Intelligence Laboratory at Cambridge, with applied research at the AWS AI lab.

Google ScholarLinkedIn

Past projects & publications

Selected research papers from our mentors' students

In their words

What our students say

  • SPARK Insights was very comprehensive. I’m leaving the programme with new skills, greater confidence, and excitement for what comes next.
    KennedyGhana
  • This course gave me insight into how to study research papers and conduct experiments independently. It also taught me the fundamentals of AI, from the mathematics behind models to how they reason and where they can be vulnerable.
    KaranUnited Kingdom
  • The programme was extremely enriching. The lectures were comprehensive, every class felt productive, and the mentors supported us throughout the process.
    SiddiqaPakistan
  • What I liked most was the structure of the programme. The mentors first explained the theory clearly, then we applied what we learned in workshops. This made complex topics much easier to understand.
    IzabellaKyrgyzstan

Have a question? Message us on WhatsApp.

Ask about start dates, whether your background is a fit, or the weekly commitment. We usually reply within a few hours.

+44 7873 168231

Prefer email? [email protected]

Start a conversationAsk about start datesRegister your interest

FAQs

Common questions

Who is this fellowship for?
Students curious about AI, mathematics, computer science, or research. It is open to Oxbridge Mentors students at school and at university, and no prior research experience is required.
What if I have not done much programming?
Some Python helps, but our pre-course material gets you up to speed before week one. If you are unsure whether your background is a fit, message us on WhatsApp and we can talk it through.
How much time does it take each week?
Around 8–10 hours a week, covering the three sessions and your own reading, coding, and experiments. Session times are agreed each week around the cohort, so it fits alongside school, a degree, or a job.
Is the programme fully remote?
Yes. Every session runs online and the cohort is coordinated on Slack. Sessions are scheduled across time zones where possible.
How big are the teams?
You work in a team of 2–4 students with a dedicated mentor.
Do students get their work published?
Most do: more than 90% of students in previous cohorts have had their work published. The write-up and submission guidance are part of the programme.
What happens if I have exams or a holiday during the eight weeks?
Usually workable. Tell us upfront so your mentor can plan around it, and expect to make up the hours either side.
Is there a certificate at the end?
Yes, on completion.
Are scholarships available?
Some scholarship support is available. Get in touch on WhatsApp or by email to discuss access support or payment options.

Next steps

Ready to start?

There is no form to fill in. Message us on WhatsApp to register your interest in the late autumn 2026 cohort, ideally before Monday 26 October 2026.

Register your interestAsk a question first
Questions? Chat with us