Pure mathematics · machine learning research
Almaty, Kazakhstan · ra312.github.io · akylzhanov.r@gmail.com
Mathematician and research engineer working at the interface of non-commutative harmonic analysis and machine learning. I focus on tabular foundation models, transformer architectures for mixed-type and structured data, and LLM evaluation—from synthetic pre-training and causal priors to benchmarking and agentic workflows.
Under what conditions do synthetic data-generating priors let a tabular transformer—trained only on synthetic distributions—generalise to unseen real-world data without task-specific fine-tuning?
At HighSky I pre-train transformers on synthetic data from Bayesian priors and structural causal graphs, aiming for strong transfer to real tabular benchmarks. In parallel I extend Hörmander–Mikhlin multiplier theory to general von Neumann algebras.
Note: Compressing long contexts with log-signatures (March 2026)
- Papers: 14 total (12 peer-reviewed + 2 preprints)
- Citations: 220+
- h-index: 7
Metrics from Google Scholar, 2026.
Full CV, talks, funding, and teaching: ra312.github.io
2024 — present · Senior Research Engineer · HighSky
- Tabular foundation models; per-feature transformers and attention for mixed-type data
- Synthetic data from structural causal graphs and diverse Bayesian priors
- LLM evaluation (MMLU, GSM8K, HellaSwag, HumanEval) and agentic multi-step reasoning
2022 — 2024 · Senior Machine Learning Engineer · Delivery Hero
- Transformer representation learning; RAG over large product catalogues
- Most Advanced Project, Global Search Domain Project Week (Jan 2023)
2020 — 2022 · Machine Learning Engineer · KCell
- Sequential modelling at scale; churn and lifetime value
- Representations for heterogeneous customer data
2018 — 2020 · Postdoctoral Research Associate (EPSRC EP/R003025/1) · Queen Mary University of London
- Harmonic analysis for Dirac-like operators on semi-finite von Neumann algebras
- Quantum group Pontryagin duality
2017 — 2018 · Research Associate (EPSRC EP/R003025/1) · Imperial College London
- Compact quantum matrix groups; Schwartz kernels and pseudo-differential calculus on compact quantum groups
Mathematics
Fourier multipliers on von Neumann algebras, compact quantum groups, Hörmander–Mikhlin theorems, non-commutative
Machine learning
Tabular foundation models, structural causal models, attention for mixed-type data, synthetic pre-training priors, RAG, LLM evaluation.
PhD, Pure Mathematics (2014–2019) — Imperial College London
Thesis:
MSc, Mathematics (2012–2014) — Eurasian National University
Specialist, Mathematics & Computer Science with Honours (2007–2012) — Lomonosov Moscow State University
Also: Vertex AI, Kubeflow, Airflow, CI/CD.
- Email: akylzhanov.r@gmail.com
- Site: ra312.github.io
- GitHub: @ra312