TC
English
Seoul, South Korea

Research engineer focused on making frontier generative models faster, smaller, and deployable at scale.

I love building things, optimizing, and solving technical problems.

View experience View projects

01

Neural network optimization with a focus on pruning and graph optimization

02

Generative AI with a focus on diffusion models, and Computer Vision

03

Various tools and frameworks: Pytorch, ONNX, etc.

Selected Impact

  1. 01

    First author on EdgeFusion, enabling on-device text-to-image generation through graph optimization

  2. 02

    Co-authored NeurIPS paper on robust curriculum learning

  3. 03

    Worked on deployment and optimization pipelines for real-world edge AI systems

  4. 04

    Specialized in pruning, graph optimization, and efficient generative models

Experience

Bridging research and deployment for efficient AI

Research engineer at Nota AI, specialising in neural network optimization.

2023 — Present Seoul, South Korea

Research Engineer · Nota AI

  • Joined the NetsPresso project as a research engineer, focusing on graph optimization.
  • First author on EdgeFusion; reduced latency enough to enable on-device text-to-image generation on mobile NPUs (in collaboration with Samsung Electronics)
  • Collaborated on various papers focusing on optimizing generative AI for edge devices.
2021 — 2022 Berlin, Germany

Researcher · Nota AI

  • Research focused on neural network optimization via pruning
2020 — 2020 Meylan, France

Research Intern · NAVER LABS Europe

  • Co-authored SuperLoss, a curriculum learning loss accepted to NeurIPS.
  • Ranked 2nd during NAVER LABS intern day for presenting research impact.

Projects

Building useful software, one project at a time

A growing index for apps, tools, experiments, and practical systems I build across whatever problem is worth solving next.

View all projects
Plume class notes page written in Markdown inside a polished native macOS interface.
01 Mac app Building

Plume

A native macOS app for beautiful local Markdown notes, everyday lists, and project management.

Free local notes and project management, without subscriptions, for people who want to keep control of their files.

macOS Markdown project management
Release updates planned

Selected Work

Professional systems and shipped AI work

01 PyTorch

NetsPresso

End-to-end platform that automates pruning, quantisation, and deployment for edge AI.

Delivered model optimisation workflows powering customer deployments across NPUs and GPUs.

PyTorch CUDA ONNX Runtime
02 Diffusion

EdgeFusion

On-device text-to-image generation system targeting mobile NPUs with optimised diffusion models.

Achieved sub-second inference budgets through graph-level optimisations and quantisation.

Diffusion TensorRT Android
03 Latent Diffusion

BK-SDM

Compressed Stable Diffusion variants for faster, lower-cost image generation inference.

Open-sourced recipe adopted by builders experimenting with efficient image generation.

Latent Diffusion FP8 Quantisation LoRA

Let’s collaborate

Thibault Castells

Questions? Feel free to reach out.

Availability

I'm open to talk about anything.