Skip to content
Nikita Pavlichenko

Cohere · Foundations/LLM pre-training/Berlin, Germany

I pre-train language models, and mostly work on what goes into them.

I'm Nikita Pavlichenko, a Member of Technical Staff on the Foundations team at Cohere. Before that I spent two and a half years at JetBrains leading Mellum, an open family of code LLMs: from the data pipeline, to training runs on 512 GPUs, to code completion that hundreds of thousands of developers use every month.

Earlier, I did research on crowdsourcing, evaluation and RLHF at Yandex and Toloka AI.

Get in touchGitHubScholarLinkedIn
  1. 1

    Yandex2018

    Search & ML internships

  2. 2

    MIPT × Inria2020

    Graph nets for protein structures

  3. 3

    Toloka AI2020

    Crowdsourcing, evaluation, RLHF

  4. 4

    JetBrains2024

    Leading Mellum code LLMs

  5. 5

    Mellum22026

    12B MoE, open weights

  6. 6

    CohereNov 2026

    Foundations team

Fig. 1·My career as a training run. The loss is made up; the checkpoints are real.

§01

Selected work

Open-weight code LLM

2026

Mellum2

A 12B mixture-of-experts model with 2.5B active parameters. I led pre-training: data mixture, filtering, dedup and FIM ablations, attention design for latency, and the runs on 512×H200.

Technical reportWeights

Production code completion

2025

Mellum in the IDE

In-IDE completion with multi-file project context. Competitive with commercial solutions in offline evals and online A/B tests, and used by hundreds of thousands of developers every month.

PaperProduct

Alignment

2023

RLHF at Toloka

Led SFT and RLHF for Toloka's LLM, from data collection to training. The model beat Falcon-40B-Instruct with a >65% win rate on real user requests. Authored the ICML 2023 tutorial on RLHF.

Open-source library

2021

Crowd-Kit

A Python toolkit for aggregating crowdsourced annotations, alongside benchmark datasets for crowdsourced audio transcription (NeurIPS) and pairwise comparisons.

PaperCode

§02

Experience

  1. CohereCurrent

    Nov 2026 – now · Berlin · remote

    Member of Technical Staff, Foundations

  2. JetBrains

    Mar 2024 – Oct 2026 · Berlin

    Staff Research Engineer / Team Lead

    • Led Mellum, JetBrains' open-source code LLM family, and a team of 7 researchers and engineers.
    • Led pre-training of Mellum2: large-scale ablations on data (mixtures, synthetic generation, filtering, dedup, FIM) and architecture (GQA/MLA, sliding-window attention), with distributed runs on 512×H200.
    • Designed and built a 10 TB+ data processing and streaming pipeline for training.
    • Shipped completion models competitive with commercial solutions offline and in online A/B tests, serving hundreds of thousands of monthly users.
    • Built large-scale RLVR pipelines for coding agents, competitive programming, instruction following and math. Owned research-engineer hiring.

  3. Yandex · Toloka AI

    Sep 2020 – Mar 2024 · Moscow · Belgrade

    AI Research Scientist

    • Led deep-learning research for Toloka AI: 6 papers on NLP, computer vision and generative AI at NeurIPS, SIGIR and HCOMP.
    • Led SFT and RLHF in Toloka's LLM project; the model beat Falcon-40B-Instruct with a >65% win rate on real user requests.
    • Authored the ICML 2023 tutorial on RLHF.
    • Built an AI-assisted annotation service that raised revenue on text-classification projects by up to 30%. 3 patent applications.

  4. MIPT & Inria NANO-D

    Mar 2020 – Sep 2020 · Moscow · Grenoble

    Research Intern

    • Developed S-GCN, spherical convolutions on molecular graphs for protein model quality assessment. State of the art on the CASP MQA challenge; published in Machine Learning: Science and Technology.

  5. Yandex

    2018 – 2019 · Moscow

    Software Engineering / ML Intern

    • Built a model for ranking video clips in web-crawler schedules that removed over 90% of deleted videos from search results.
    • Sped up all search anti-spam systems by 5% with the fastest C++ hash table at Yandex.

Education

  • MSc, Computer Science · Moscow Institute of Physics and Technology2021 – 2023
  • BSc, Computer Science · Moscow Institute of Physics and Technology2017 – 2021

Toolbox

Languages
Python, C++, Rust, Bash
Training
PyTorch, multi-node mixed precision, MoE, LLM evaluation
Data & infra
Spark, AWS, Kubernetes, Docker, multi-TB pipelines

§03

Publications

Google Scholar
  1. [11]

    Mellum2 Technical Report

    M. Kojic, I. Bondyrev, A. de Moor, J. Shtok, P. Borovlev, K. Lysaniuk, M. Kannan, I. Dolgov, N. Pavlichenko

    arXiv:2605.31268· 2026

  2. [10]

    Mellum: Production-Grade in-IDE Contextual Code Completion with Multi-File Project Understanding

    N. Pavlichenko, I. Nazarov, I. Dolgov, E. Garanina, D. Ustalov, I. Bondyrev, K. Lysaniuk, E. Vu, K. Chekmenev, J. Shtok, Y. Golubev, A. Semenkin, U. Sazanovich

    arXiv:2510.05788· 2025

  3. [9]

    Best Prompts for Text-to-Image Models and How to Find Them

    N. Pavlichenko, D. Ustalov

    SIGIR 2023· 2023

  4. [8]

    WSDM Cup 2023 Challenge on Visual Question Answering

    D. Ustalov, N. Pavlichenko, D. Likhobaba, A. Smirnova

    WSDM Cup 2023· 2023

  5. [7]

    CrowdSpeech and Vox DIY: Benchmark Dataset for Crowdsourced Audio Transcription

    N. Pavlichenko, I. Stelmakh, D. Ustalov

    NeurIPS 2021 Datasets and Benchmarks· 2021

  6. [6]

    Spherical Convolutions on Molecular Graphs for Protein Model Quality Assessment

    I. Igashov, N. Pavlichenko, S. Grudinin

    Machine Learning: Science and Technology· 2021

  7. [5]

    IMDB-WIKI-SbS: An Evaluation Dataset for Crowdsourced Pairwise Comparisons

    N. Pavlichenko, D. Ustalov

    NeurIPS 2021 Data-Centric AI Workshop· 2021

  8. [4]

    Learning from Crowds with Crowd-Kit

    D. Ustalov, N. Pavlichenko, B. Tseitlin

    arXiv:2109.08584· 2021

  9. [3]

    A General-Purpose Crowdsourcing Computational Quality Control Toolkit for Python

    D. Ustalov, N. Pavlichenko, V. Losev, I. Giliazev, E. Tulin

    HCOMP 2021 Works-in-Progress & Demos· 2021

  10. [2]

    VLDB 2021 Crowd Science Challenge on Aggregating Crowdsourced Audio Transcriptions

    D. Ustalov, N. Pavlichenko, I. Stelmakh, D. Kuznetsov

    Crowd Science Workshop at VLDB 2021· 2021

  11. [1]

    Toloka at LMNL Challenge: Learning from Noisy Labels with Annotator Reliabilities and Self-Supervised Training

    B. Tseytlin, N. Pavlichenko, D. Likhobaba, A. Smirnova, D. Ustalov

    LMNL Challenge

TalksTutorial on reinforcement learning from human feedback, ICML 2023.

§04

Writing

All posts →

§05

Contact

Email is the best way to reach me. I'm always happy to talk about pre-training data, evals, and making experiments cheaper.

nikita.v.pavlichenko@gmail.com