Erwan Zerhouni
Principal Machine Learning Engineer at Cisco
Zurich, Zurich, Switzerland
Summary
🤩
Rockstar🎓
Top SchoolErwan Zerhouni is a Principal Machine Learning Engineer and seasoned leader based in Zurich, bringing over a decade of experience shipping large-scale ML products, notably driving Automatic Speech Recognition for Cisco Webex. He combines hands-on research and engineering—authoring production-grade multilingual ASR systems, end-to-end model pipelines, and cost- and time-saving optimizations that reduced inference costs by over 90% and training time by 75%. Erwan has led cross-functional teams to deploy billions of processed minutes per year, while mentoring engineers and serving as an internal patent reviewer. His background spans academic computer vision and bioimage analysis at IBM Research to practical cloud-native ML at Cisco, reflecting both deep algorithmic expertise and system design skills. An active contributor to open-source ASR tooling, he implemented fast beam search and streaming optimizations in the k2-fsa/sherpa speech-to-text framework, bridging state-of-the-art models with production decoding.
5 years of coding experience
9 years of employment as a software developer
Master of Science (MSc), Mathematics, Machine Learning and Computer Vision (MVA), Master of Science (MSc), Mathematics, Machine Learning and Computer Vision (MVA) at Ecole Centrale ParisINSA RennesLycée DescartesMaster of Engineering (MEng), Télécommunication, Services et Usages, Master of Engineering (MEng), Télécommunication, Services et Usages at Institut national des Sciences appliquées de Lyon / INSA LyonEngineer's degree, Computer Science, Exchange Student, Engineer's degree, Computer Science, Exchange Student at Yonsei University
French, English, Spanish, Arabic, German
Github Skills (11)
speech-to-text10
pytorch10
python10
asr10
cprogramming-language9
c-language9
cpp9
websockets-client8
websocket8
client-websocket8
websocket-client8
Programming languages (4)
Github contributions (5)
k2-fsa/sherpa
Jul 2022 - Oct 2022
Speech-to-text server framework with next-gen Kaldi
Role in this project:
Back-end Developer
Back-end DeveloperContributions:36 reviews, 21 commits, 11 PRs in 2 months
Contributions summary:Erwan implemented fast beam search functionality within the speech-to-text framework, specifically for the stateless Emformer and Conformer models. This involved adding new decoding methods, including "fast_beam_search", and modifying existing code to incorporate these methods, along with associated parameters like beam size and context settings. Their work extended to optimizing the streaming server's capabilities by adding features such as timestamps and tokens to aid in the decoding process. The user also refactored and updated code across several files related to streaming ASR and offline model implementations.
pytorchtransducerkaldicpppython
ezerhouni/icefall
Jun 2022 - Mar 2024
Contributions:9 PRs, 74 pushes, 13 branches in 1 year 9 months
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.