PeopleofSuperintelligence

Builder

Mickaël Seznec

Also known as Mickael Seznec

Mickaël Seznec builds GPU systems for AI, from embedded vision to large language models. His 2021 thesis at CentraleSupélec optimized image convolution and optical flow for embedded GPUs, and he co-authored a 2022 article on the im2tensor algorithm. He co-authored SmoothQuant for 8-bit language-model quantization and Mistral papers on reasoning, audio and coding models in 2025. In February 2026, he spoke for Mistral at a Paris meetup on Mistral Large 3 deployment strategies.

Works
14

Publication numbers from OpenAlex, as of 22 October 2026. Citations and h-index are hidden because this OpenAlex record may be incomplete or mix in other authors.

Timeline

  1. 2021

    A thesis on embedded GPUs

    At CentraleSupélec's L2S laboratory, his public thesis defense is scheduled for October 2021 under Nicolas Gac, with François Orieux as co-supervisor. The work modifies image convolution to use GPU matrix units, doubling performance for large kernels, and runs optical flow at 60 frames per second on a 30-watt embedded platform.

  2. 2022

    The im2tensor algorithm

    In 2022, he coauthors the Journal of Real-Time Image Processing article “Computing Large 2D Convolutions on GPU Efficiently with the im2tensor Algorithm.”

  3. 2022–2023

    SmoothQuant for LLM quantization

    In November 2022, he coauthors SmoothQuant with Guangxuan Xiao, Ji Lin, Hao Wu, Julien Demouth and Song Han, a training-free post-training quantization method for large language models. It enables 8-bit weights and activations, with up to 1.56× speedup and 2× memory reduction; ICML publishes the paper in 2023.

  4. June 2025

    Magistral and pure reinforcement learning

    In June 2025, he is one of 99 authors of Magistral, Mistral's first reasoning model, and its scalable reinforcement-learning pipeline built on Mistral's own models and infrastructure. Magistral Medium is trained on Mistral Medium 3 with reinforcement learning alone, and Magistral Small is released under Apache 2.0.

  5. July 2025

    Voxtral audio models

    In July 2025, he coauthors Voxtral, which presents Voxtral Mini and Voxtral Small as multimodal audio chat models released under the Apache 2.0 license. Voxtral Mini builds on Ministral 3B, Voxtral Small on the Mistral Small 3.1 24B backbone, and a 32K context window lets them handle audio up to 40 minutes.

  6. August 2025

    The Devstral coding model

    The Devstral technical report from Mistral AI and All Hands AI appears on arXiv in August 2025, with him among its authors. It presents Devstral-Small as a lightweight open-source 24B model for code agents.

  7. November 2025

    GPU engineer at Mistral AI

    At Mistral AI, he is a GPU engineer and speaks with Mathis Felardos on “P/D Disaggregated Serving in Production with vLLM” at Europe's first official vLLM meetup in Zürich in November 2025. Red Hat, IBM and Mistral AI host the event.

  8. February 2026

    Mistral Large 3 deployment

    In February 2026, he speaks for Mistral alongside NVIDIA's Julien Debache at the NVIDIA Open Source Developer Meetup in Paris, presenting deployment strategies for Mistral Large 3.

Career

  1. NowMistral AIGPU Engineer

Education

  1. Université Paris-SaclayPhD

From the biography.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Mistral AI. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.
  2. Companies supply the most prior-work hires at 7 of the 9 frontier AI labsIncludes Mistral AI. Our estimates of the organizations where 550 people worked directly before joining 9 frontier AI labs across 620 career transitions, comparing the main feeder organizations per lab.
  3. Of 943 people shaping AI who worked at Google, 446 work there nowIncludes Mistral AI. Our estimates of where the 943 people shaping AI who worked at Google and Google DeepMind work now: 446 at Google and 452 across 278 other organizations.
  4. 5,000 frontier users would be 38.2% of the core labs' 13,100 reported staffIncludes Mistral AI. 5,000 frontier users would be 38.2% of the core labs' 13,100 reported staff. We test the estimate against 3,913 named model contributors and dated all-staff records across Google DeepMind, OpenAI and Anthropic, plus six additional frontier-grade labs.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Mickaël Seznec yet. Send us a free-licensed photo.

  1. Avis de Soutenance Monsieur Mickaël SEZNEC | L2S UMR 8506 l2s.centralesupelec.fr. Checked 21 October 2026.
  2. CV HAL - Nicolas Gac cv.hal.science. Checked 21 October 2026.
  3. [2211.10438] SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models arxiv.org. Checked 21 October 2026.
  4. SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models proceedings.mlr.press. Checked 21 October 2026.
  5. [2506.10910] Magistral arxiv.org. Checked 21 October 2026.
  6. [2507.13264] Voxtral arxiv.org. Checked 21 October 2026.
  7. Voxtral arxiv.org. Checked 21 October 2026.
  8. [2509.25193] Devstral: Fine-tuning Language Models for Coding Agent Applications arxiv.org. Checked 21 October 2026.
  9. Devstral: Fine-tuning Language Models for Coding Agent Applications arxiv.org. Checked 21 October 2026.
  10. Zürich vLLM Meetup · Luma luma.com. Checked 21 October 2026.
  11. NVIDIA Open Source Developer Meetup · Luma luma.com. Checked 21 October 2026.