Nikita Kazeev

Curriculum Vitae

Nikita Kazeev

Download PDFOctober 2026

Research Scientist at the intersection of AI and Physics. Postdoc with Kostya Novoselov at NUS.

12

years in research

9

students mentored

$3.4M

grant co-PI

20+

conference talks

Breakthrough Prize

Breakthrough Prize 2025

Skills

  • Information extraction & retrieval — pipelines that turn unstructured web pages and PDFs into structured, queryable data; LLM deep-research agents, embedding-based semantic matching, and grounding / hallucination detection.
  • Generative modelling for scientific data — GANs, autoregressive transformers, and diffusion; from fast detector simulation to symmetry-aware crystal generation.
  • Agent Reliability & UQ — developed geometric and statistical frameworks to predict LLM agent reasoning failures using semantic abstraction trajectories.
  • Aspirational Alignment & Verification — formulated alignment strategies based on high-level human values and established benchmarks (e.g. Foresight-Phys) for validating AI systems in complex environments.
  • Deep learning for structured data — designed and trained GANs, autoregressive transformers, diffusion models, and graph neural networks end-to-end in PyTorch.
  • Uncertainty quantification — built the first methods for estimating the uncertainty of conditional GANs; calibrated generative and discriminative models.
  • ML on structured & tabular data — from gradient boosting (CatBoost) to transformers; graph and atomistic representations.
  • Inference and systems at scale — C++ production integration deployment, large-scale compute on HPC clusters, and reproducible experiment pipelines.
  • Physics — high-energy and condensed-matter physics
  • Research leadership — lead multidisciplinary teams, chair conferences, and mentor students from idea to publication.

Work Experience

National University of Singapore

Postdoc under [Kostya Novoselov](https://en.wikipedia.org/wiki/Konstantin_Novoselov)

2022–present

Foresight-Phys: papers into forecastable data
  • Built an extraction pipeline that ingests scientific PDFs from public URLs and parses them into structured, queryable JSON — LLM structured-output schemas plus an automatic suitability-filtering pass
  • Engineered for throughput, correctness, and observability --- parallel model calls with retry/backoff, a concurrency-safe shared cache, and Langfuse tracing capturing the full experiment lineage (input, predictions, metrics, corrections)
  • Scored predictions with proper scoring rules (CRPS, Brier) and calibrated per-field uncertainty; released as the Foresight-Phys benchmark
AI4X conference operations, rebuilt as an engineering system
  • Turned the Program Chair appointment — normally hand-run administration — into an engineering system: a pipeline orchestrating deep-research and LLM APIs that reads the open web and returns structured speaker profiles, with routing, fallback, and tracing
  • Built embedding-based matching and CP-SAT scheduling to place 100+ speakers into ranked, thematically-coherent sessions
Wyckoff Transformer, a generative model for materials
  • Cut inference to 0.05 GPU-ms/structure — 4 orders of magnitude faster than prior models
  • Designed a coordinate-free, permutation-invariant autoregressive Transformer encoder that represents crystals by symmetry, improving materials-discovery yield by 1.24×
  • Implemented the model in PyTorch; shipped as a production-ready package on PyPI with GitHub Actions CI, YAML-config-driven experiments, and reproducible evaluation for all paper metrics
Leadership
LLM Agent Safety & Alignment
  • Conceived and developed a novel geometric framework to predict multi-step reasoning failures in LLM agents using semantic abstraction trajectories
  • Designed uncertainty quantification (UQ) algorithms for LLM reasoning loops, improving agent error detection and calibration
  • Developed and open-sourced reference-checker, a tool for verifying claims and citation references in scientific papers to detect and mitigate LLM hallucinations
  • Authored a position paper on Pluralistic Alignment, advocating for aligning AI models with aspirational human values rather than replicating behavioral flaws
Wyckoff Transformer, a generative model for materials
  • Developed a coordinate-free permutation-invariant autoregressive Transformer encoder model for material generation and property prediction based on symmetry inductive bias
  • Increased materials discovery yield by 1.24×
  • Cut inference to 0.05 GPU-ms/structure — 4 orders of magnitude faster than prior models
  • Implemented the model in PyTorch; shipped as a production-ready package on PyPI with GitHub Actions CI, YAML-config-driven experiments, and reproducible evaluation code for all paper metrics
  • Orchestrated large-scale (10k structures) DFT computations with VASP on an HPC cluster
  • Led a team of 6
Wyckoff Transformer, a generative model for materials
  • Alignment by design: enforced physical principles directly at the architectural level as hard constraints, mathematically guaranteeing plausible generations
  • Solved a systemic failure mode where generative models default to asymmetric outputs despite symmetric training data, mitigating this structural bias via coordinate-free representations
  • Increased materials discovery yield (S.U.N.) by 1.24× while cutting inference time by 4 orders of magnitude (to 0.05 GPU-ms per structure)
  • Orchestrated large-scale (10k structures) DFT computation loops to verify out-of-distribution stability; led a team of 6
ML for defects in 2D crystals
  • Conceived and implemented a novel sparse representation of crystals with defects
  • Cut energy-prediction error 3.7× vs. the best prior model
  • Optimised training and inference: 4× less memory, 8× fewer GPU operations
  • Built a reproducible data-preparation pipeline: DVC-tracked versioning from raw data through feature extraction to the train/test splits, W&B experiment tracking with per-fold config logging, containerised environments (Docker+Poetry)
  • Led a team of 3
Cross-model evaluation benchmark for crystal generative models
  • Created and published an open benchmark dataset spanning 9 generative models for crystal structure prediction --- solicited and curated contributions from the authors of each model (DiffCSP, WyCryst, CrystalFormer, MatterGen, and others) across research groups into a common format, and computed DFT validation for a subset
  • Co-authored the LeMat-GenBench benchmark paper (NeurIPS 2025 Workshop), contributing the dataset and analysis of evaluation metrics for comparing generative models of crystalline materials

Perfomax

AI Advisor

2026–present

Constructor Group

Machine Learning Consultant

2021–present

Most engagements here are under NDA

reference-audit: matching citations to the real documents they point to
  • Resolves every citation in a paper to the real document it points to, and flags references that match nothing real (likely hallucinations)
  • Retrieves candidates from seven scholarly databases (Crossref, OpenAlex, Semantic Scholar, arXiv, DBLP, Open Library, Google Books), entity-resolves duplicate and preprint/published records into one work, and ranks them with interpretable features — sending only ambiguous cases to an LLM
  • Crawls and parses cited web pages directly, re-rendering JavaScript single-page apps in a headless browser so an empty app shell is never misread as a mismatch

⟨CERN ∣ Yandex → HSE University⟩

Researcher

2014–2022

2025 Breakthrough Prize in Fundamental Physics as a member of LHCb.

CatBoost, an open-source gradient-boosting library
  • Co-authored the original 2017 CatBoost preprint
  • Did the experiments proving CatBoost's ordered boosting indeed removes the gradient estimation bias
  • Ran benchmarks across Yandex's compute cluster --- automated parallel execution, dependency management, and result collection
Muon identification at the LHCb experiment at CERN
  • Extracted signal from noisy, tabular detector data under latency constraints
  • Built a CatBoost model that cut the false-positive rate ~30% in the low-momentum region
  • Shipped the model into the LHCb trigger --- a production C++ & Python system which processes data at 30 MHz | 5 TB/s (Run 3 detector readout/HLT1 input)
  • Packaged the work as a data science competition problem for IDAO-2019
Machine learning on noisy data
  • Derived a statistically rigorous method to train classifiers on sPlot-weighted (label-noisy) data
Generative models uncertainty estimation
  • Developed the first methods for estimating uncertainty of conditional GANs
  • Led a team of 3 students
Generative models for fast simulation
  • Built a GAN for high-fidelity, tabular-data simulation of a Cherenkov detector — ~10⁵× faster than the full Geant4 simulation
CatBoost: Combatting Biases in Boosting
  • Designed and built distributed experiment infrastructure to scale systematic bias analysis across large-scale tabular datasets
  • Investigated dynamic boosting formulations to eliminate target leakage and prediction shift, mitigating statistical bias in gradient boosting

Education

HSE University & Sapienza Università di Roma

2016–2020

PhD in Computer Science and Physics (Double Degree)

Supervisors: Andrey Ustyuzhanin & Barbara Sciascia

Thesis: Machine Learning for particle identification in the LHCb detector

Product Management course at Yandex

2018

Focused coursework in product strategy and execution.

Yandex School of Data Analysis

2013–2015

Master's-level CS programme

Algorithms, machine learning, deep learning, and distributed systems.

Moscow Institute of Physics and Technology

MS in Physics

2014–2016

Optimisation of data processing of the LHCb experiment.

BS in Physics

2010–2014

Wave-packet molecular dynamics of electrons in nonideal plasma.

International Junior Science Olympiad (IJSO)

Korea, 2008

Silver medal

Mentorship

  • Mentored 9 students (1, 2, 3, 4, 5, 6, 7, 8, 9), 3 interns, and student workshop projects (1, 2).

Teaching & Outreach

Selected Publications

Full list on Google Scholar →
  1. Kazeev, Nikita, and Ian Babich. Foresight-Phys: A Benchmark for Forecasting the Results of Physical Experiments. Forecasting as a New Frontier of Intelligence Workshop at ICML 2026.
  2. Kazeev, Nikita, and Phan Bui Nhat Huyen. Position: Align AI to Our Aspirations, Not Our Flaws. Pluralistic Alignment Workshop at ICML 2026.
  3. Kazeev, Nikita, and Andrey Ustyuzhanin. The Geometry of Reasoning Failure: Predicting Agent Errors from Semantic Scale Trajectories. Statistical Frameworks for Uncertainty in Agentic Systems Workshop at ICML 2026.
  4. Duval, Alexandre, Siddharth Betala, Samuel P. Gleason, Andy Xu, Georgia Channing, Daniel Levy, Ali Ramlaoui, Clémentine Fourrier, Chaitanya K. Joshi, Nikita Kazeev, Sékou-Oumar Kaba, Félix Therrien, Alex Hernández-García, Rocío Mercado, and N M Anoop Krishnan. LeMat-GenBench: Bridging the gap between crystal generation and materials discovery. AI for Accelerated Materials Design --- NeurIPS 2025, 2025.
  5. Kazeev, Nikita, Wei Nong, Ignat Romanov, Ruiming Zhu, Andrey E Ustyuzhanin, Shuya Yamazaki, and Kedar Hippalgaonkar. Wyckoff Transformer: Generation of Symmetric Crystals. Proceedings of the 42nd International Conference on Machine Learning, 2025.
  6. Kazeev, Nikita, Abdalaziz Rashid Al-Maeeni, Ignat Romanov, Maxim Faleev, Ruslan Lukin, Alexander Tormasov, A. H. Castro Neto, Kostya S. Novoselov, Pengru Huang, and Andrey Ustyuzhanin. Sparse representation for machine learning the properties of defects in 2D materials. npj Computational Materials 9, 113 (2023).
  7. Anderlini, L., F. Archilli, A. Cardini, V. Cogoni, M. Fontana, G. Graziani, N. Kazeev, H. Kuindersma, R. Oldeman, M. Palutan, M. Santimaria, B. Sciascia, P. De Simone, and R. Vazquez Gomez. Muon identification for LHCb Run 3. 15.12, T12005 (2020). Alphabetic order
  8. Borisyak, Maxim, and Nikita Kazeev. Machine Learning on data with sPlot background subtraction. Journal of Instrumentation 14.08 (2019). Alphabetic order, I'm the corresponding author.
  9. Derkach, Denis, Nikita Kazeev, Fedor Ratnikov, Andrey Ustyuzhanin, and Alexandra Volokhova. Cherenkov detectors fast simulation using neural networks. Nuclear Instruments and Methods in Physics Research Section A (2019). Alphabetic order, I'm the corresponding author.

Service

  • Reviewer for NeurIPS, RSC Advances, Machine Learning: Science and Technology, Digital Discovery
  • Peer Staff Supporter at NUS, serving as a first-line support for mental wellbeing.