Back to home

Isaac Bautista Silva

B.S. in Physics Engineering

Education

Tecnológico de Monterrey Campus Monterrey

B.S. Physics Engineering

Monterrey, Mexico

07.2018 - 04.2023

Technische Universität Dresden

B.S. Physics (Exchange) – Funded by German Academic Exchange Service (DAAD)

Dresden, Germany

07.2021 - 03.2022

TheResidency

Accelerator Program for Founders and Researchers

San Francisco, USA

01.2024 - 05.2024

Significant Courses: Mathematical Physics, Solid State Physics, Linear Algebra, Numerical Methods, Physics Computing, Quantum Physics, Parallel and High Performance Computing (Germany), Machine Learning (Germany)

Work Experience

Porsche

Stuttgart, Germany

Machine Learning Intern

04.2022 – 08.2022

  • Developed and deployed deep-learning models for Porsche's next-generation electric vehicle platform, taking projects from proof of concept to on-car deployment.
  • Designed and validated multi-sensor tracking / localization algorithms that improved fleet-level position accuracy and robustness under high-speed scenarios.
  • Led the over-the-air pipeline for automatic model updates, ensuring every prototype received fresh weights after the best training rounds.

Zepheon [link]

Remote

Founder

05.2024 – 05.2025

  • Launched and scaled a ML-consulting studio that partners with seed-stage startups to turn research ideas into shipped products.
  • Completed 14 different AI projects, ranging from training and distilling models for faster inference to enterprise level Agents to developing RAG pipelines for tens of thousands of legal documents.
  • Delivered full-stack solutions: from training or fine tuning the models to APIs in .NET and FastAP, all the way to clean React interfaces on the front end.

Projects

miniSAM, Segment Anything Model GPU-Free [link]

2025

  • Distilled Meta's SAM: built a one-command teacher-student pipeline that shrank the 2.4B param ViT to a 5 M TinyViT while holding IoU within limits, trainable on a single RTX 4090.
  • Engineered a zero-server ONNX runtime that handles all preprocessing and inference entirely in the browser, delivering very fast inference and GPU-free segmentation on laptops and mid-range phones.
  • Developed production ready npm packages (minisam, minisam-react) with typed hooks and plug-and-play UI; all with live demos, open-sourced end-to-end.

Diffusion Math from Scratch, Video Lectures [link]

2025

  • Derived the mathematics behind Diffusion from absolute first principles, in a series of lectures with each one of the concepts explained by hand.
  • Recreated all of the individual papers from the original idea to state of the start: Autograd Engine, UNet, DDPMs, Improved DDPMs, VAEs, Attention, etc. All code available online.
  • Developed my own educational-tech software [link] to teach and test all of the concepts in Diffusion. It has a self hosted backend that runs text-to-image, inpainting, segmentation, ControlNets, etc.