# Egor Lakomkin

> Research Scientist | Multimodal & Speech Foundation Models

Egor Lakomkin is a Member of Technical Staff at Gradium in Paris. Previously, he was a Staff Research Scientist at Meta Superintelligence, an Applied Scientist at Amazon Alexa, and a Doctoral Researcher at the University of Hamburg.

## Primary links

- Website: https://egorlakomkin.github.io/
- CV: https://egorlakomkin.github.io/files/ResumeEgorLakomkin.pdf
- Google Scholar: https://scholar.google.com/citations?user=Svmv-7EAAAAJ&hl=en
- LinkedIn: https://www.linkedin.com/in/egor-lakomkin-1032b337
- GitHub: https://github.com/EgorLakomkin
- Email: mailto:egor.lakomkin@gmail.com

## Research focus

- Multimodal and speech foundation models
- Streaming speech LLMs
- Tokenization and large-scale pretraining
- Reinforcement-learning post-training
- Model evaluation, data quality, and distributed GPU training

## Selected publications

- Efficient Streaming LLM for Speech Recognition (ICASSP 2025): https://arxiv.org/abs/2410.03752
- The Llama 3 Herd of Models (2024): https://ai.meta.com/research/publications/the-llama-3-herd-of-models/
- AudioChatLlama: Towards General-Purpose Speech Abilities for LLMs (NAACL 2024, Long Papers): https://aclanthology.org/2024.naacl-long.309/
- End-to-End Speech Recognition Contextualization with Large Language Models (ICASSP 2024): https://arxiv.org/abs/2309.10917