Egor Lakomkin
Research Scientist at Gradium
Paris, Ile-de-France
Summary
🤩
Rockstar🎓
Top SchoolEgor Lakomkin is a research scientist with 12 years of experience building and deploying speech and multimodal language models, currently focused on voice AI and online reinforcement learning at Gradium after leading related efforts at Meta Superintelligence Labs. He blends deep academic training (PhD in AI) with production experience from Amazon Alexa, where he worked on streaming end-to-end ASR, and notable open-source contributions to DeepSpeech.pytorch improving dataset support and model flexibility. His recent work centers on full-duplex speech models, speech tokenization, and aligning large language models to understand audio, aiming to make conversational agents more steerable and interactive. He has a track record of latency-focused solutions—from low-latency emotion recognition research to microsecond trading systems—demonstrating a habit of optimizing models for real-time use. Based in Paris, he brings both research rigor and hands-on engineering across Python and PyTorch ecosystems. An interesting thread through his career is repeatedly turning research prototypes into production-grade systems, including a consumer-facing Android app and scalable Alexa components.
12 years of coding experience
12 years of employment as a software developer
Doctor of Philosophy - PhD, Artificial Intelligence, Doctor of Philosophy - PhD, Artificial Intelligence at University of HamburgNon-degree program, Computer Linguistics, Non-degree program, Computer Linguistics at Higher School of EconomicsMaster's Degree, Computer Science, Master's Degree, Computer Science at Bauman Moscow State Technical University
Stackoverflow
1,424reputation
116kreached
18answers
14questions
pytorch
top-5%
Github Skills (20)
pytorch10
python10
machine-learning10
rnn-model10
n10
data-preprocessing10
data-loading10
speech-recognition10
data-augmentation9
sox8
computer-vision6
convolution6
neural-network6
deep-learning6
tensor6
Programming languages (7)
Github contributions (5)
SeanNaren/deepspeech.pytorch
Apr 2017 - Dec 2017
Speech Recognition using DeepSpeech2.
Role in this project:
ML Engineer
ML EngineerContributions:20 commits, 11 PRs, 41 comments in 8 months
Contributions summary:Egor significantly contributed to the data processing and model training aspects of the DeepSpeech.pytorch repository. They integrated support for multiple datasets including TED-LIUM, VoxForge, and LibriSpeech, writing scripts to download, preprocess, and create manifests for these datasets. Furthermore, the user refactored the model to accept different RNN types (simple_rnn, gru, lstm) and also added bucketing to improve the training process.
deepspeech2recognitionspeechspeech-recognition
EgorLakomkin/KTSpeechCrawler
May 2018 - Feb 2020
Automatically constructing corpus for automatic speech recognition from YouTube videos
Contributions:15 commits, 14 pushes, 1 branch in 1 year 8 months
videoscorpusrecognitionspeech-recognitionmachine-learning
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.