Personal Statement — UK Global Talent Visa (Exceptional Talent)

Daniil Merkulov

---

My work sits at the intersection of mathematical optimisation and artificial intelligence — an area that has become the decisive battleground for technological leadership over the past decade. I am a researcher, educator, and builder: I design optimisation methods that make large AI systems trainable, I teach these ideas to the next generation of engineers, and I build AI agents that put these methods into practice. I am applying for the Global Talent Visa because the United Kingdom is where I intend to do my most consequential work.

Research and academic contributions

My doctoral research, completed at Lomonosov Moscow State University under the supervision of Professor Ivan Oseledets — one of the world’s leading figures in tensor methods for machine learning — addresses convergence theory and practical efficiency of first- and second-order optimisation algorithms for neural networks. The resulting thesis (222 pages) synthesises original theoretical results with numerical experiments across large-scale settings. The methods I developed address problems that any serious AI laboratory encounters daily: how to train bigger models faster, with tighter convergence guarantees.

Alongside the thesis, I have published peer-reviewed work on gradient methods, stochastic optimisation, and operator-theoretic approaches to machine learning problems. My academic profile spans pure convergence analysis, applied numerical methods, and the bridge between the two — the kind of cross-disciplinary depth that produces reusable insight rather than one-off results.

Teaching and knowledge transfer at national scale

From 2019 to 2024, I served as a senior lecturer and course lead at the Moscow Institute of Physics and Technology (MIPT), where I designed and delivered what became the Institute’s flagship optimisation course for students of machine learning. Over five years this course reached several thousand students. The open materials — lecture notes, problem sets, Python notebooks — are used by independent learners and other universities across the Russian-speaking world.

I simultaneously ran an equivalent programme at the Higher School of Economics (HSE), Faculty of Computer Science, one of Russia’s top-ranked departments for data science. Teaching at both institutions in parallel required me to make difficult pedagogical choices: what is genuinely fundamental versus what is fashion, and how to communicate hard mathematics to practitioners without losing rigour. I believe this is one of the rarest and most exportable skills in AI education.

Beyond the university setting, I was invited to give lectures to senior executives at major corporations and to government officials across multiple levels. These were not simplified talks: the audiences were decision-makers who needed to understand what AI can and cannot do. Translating frontier research into decisions is a different skill from teaching or publishing, and developing it has made me a more complete contributor.

Industry leadership: AI agents at scale

At Sber’s AI4Science division — one of Russia’s largest corporate AI research groups — I led the AI-agents team. We built systems that combined language models, retrieval, and tool use to automate complex scientific workflows. This work was shipped into internal pipelines affecting thousands of researchers and analysts. Leading an applied research team taught me the gap between academic elegance and production reliability, and how to close it.

This experience also convinced me that the next phase of AI development will be dominated by teams that understand both the mathematics of learning and the engineering of systems that act in the world. I want to be part of building that infrastructure.

Why the United Kingdom

The UK has a specific combination of assets that I do not find elsewhere at this concentration: world-class universities with strong optimisation and AI groups (Cambridge, Oxford, Imperial, UCL, Edinburgh), a deep-tech ecosystem that is willing to translate research into products, and a tradition of rigorous, open scientific culture.

Concretely, I intend to:

1. Continue research on optimisation methods for foundation models — specifically, adaptive and variance-reduced methods for sparse and structured models, where theoretical gaps remain wide and practical impact is immediate.
2. Build open educational infrastructure for AI optimisation that matches the best international standards — in English, accessible globally. The UK is the right base from which to build something with genuine international reach.
3. Collaborate with UK industry on AI systems that require reliable optimisation at inference time (agents, planners, multi-step reasoners) — a category that will grow sharply over the next five years.

My academic record, my track record of delivering education at scale, and my experience leading applied AI work give me a concrete foundation for each of these goals. I am not arriving with a vague ambition to “work in AI”; I am arriving with a specific research programme, a teaching methodology, and demonstrated ability to execute.

Summary

I am a researcher who has spent a decade making optimisation methods work — in theory, in the classroom, and in production. The problems I work on are exactly the ones that will shape what AI systems can do in the next decade. I want to do that work in the United Kingdom, where the scientific environment, the talent pool, and the technology ecosystem make it possible to do it at the highest level.

---

[ЧЕРНОВИК — требует вычитки Данилой: уточнить названия публикаций, добавить конкретные paper titles/venues если есть, проверить даты работы в AI4Science и МФТИ, убрать/заменить любые факты, которые не хочется светить в официальном заявлении. Объём: ~870 слов — в лимит 1000 слов входит.]