About

About

Hi there, I’m Sebastià Agramunt Puig, a Software & AI Engineer based in the San Francisco Bay Area (California). I grew up in a small mediterranean town not too far from Barcelona where I went to college. I studied Physics at Universitat Autónoma de Barcelona and continued my academic life with a PhD in theoretical electromagnetism in the same university, working on magnetic levitation with superconductors and magnetic recording with nanoscale magnets (download dissertation here). Along the way I’ve authored 12 peer-reviewed publications and hold 1 patent — you can find them all on ORCID.

My interests are wide but there are always two common denominators, software development and mathematics. Over more than ten years in industry I’ve worked on routing algorithms, privacy-preserving machine learning, and high-performance computing, comfortable both leading small teams and working independently. I currently work at Eikon Therapeutics as a Staff Software Engineer, where I write CUDA algorithms for protein detection and localization in high-throughput drug screening. I’m particularly interested in backend engineering, high-performance computing, CUDA programming, and LLM inference, and I’m always happy to talk about opportunities in those areas.

The blog

Over the years I’ve been keeping notes about topics I like to learn and have been useful in my professional life. The idea for this blog is to keep a log for myself as well as share with other people that may have similar interests to mine. Please, reach out if you have questions about the content or find errors in the posts.

Resume

Download Resume (PDF)

Experience

Staff Software Engineer Current

Eikon Therapeutics · Millbrae, CA Jan 2021 – Present
  • Implemented protein detection and localization algorithms in pure CUDA, with Python bindings and CI/CD for x86 and ARM, achieving a 100x speedup processing 1.5GB movies.
  • Contributed to building the data processing pipeline for single-molecule tracking, including image preprocessing algorithms, PostgreSQL database design, and distributed computing.
  • Trained and served image segmentation models (U-Net).
  • Contributed CI/CD improvements across the company's entire stack and advocated for best practices in testing and artifact publishing.
  • Implemented the physics simulation of Brownian motion for proteins in an in-house simulation tool.

Privacy Preserving AI Researcher Contractor (Remote)

Neurocat · Berlin, Germany Jan 2021 – Jun 2021
  • Coauthored a report on the state of the art in privacy-preserving machine learning with Germany's Federal Office for Information Security.
  • Focused research on threats associated with transfer learning, including backdoor and adversarial poisoning attacks.

AI Engineer and Privacy Preserving Machine Learning Lead

Telefónica Alpha · Barcelona, Spain Nov 2018 – Aug 2020
  • Initiated and led the company's Privacy Preserving Machine Learning initiative from scratch.
  • Acquired a deep understanding of the mathematics behind Differential Privacy, Secure Multi-Party Computation (SMPC), and Fully Homomorphic Encryption (FHE).
  • Developed a proof-of-concept for Federated Learning with Secure Aggregation (an SMPC technique) and led a small team to evolve it into a functional mobile product.
  • Refactored code to transition a neural collaborative filtering recommender system into production.

Research Scientist, Privacy in Machine Learning

OpenMined · Remote Mar 2020 – Aug 2020
  • Developed an "Introduction to Cryptography" MOOC covering mathematical foundations and Python implementations, reaching over 7,000 students worldwide.
  • Researched secure inference on secretly shared machine learning models, focusing on activation function approximation within algebraic rings.

AI & Routing Algorithms Engineer

Shotl · Barcelona, Spain Jul 2016 – Oct 2018
  • Designed and implemented the company's core routing algorithm from scratch, combining simulated annealing, depth-first search, and Dijkstra's algorithm, following an extensive literature review.
  • Built fast, reliable software to deploy the routing algorithm in production.
  • Analyzed urban demand patterns using machine learning techniques.
  • Developed a simulator to evaluate the performance of the routing algorithms.

Data Science Consultant

Accenture Digital · Sant Cugat (Barcelona), Spain Jul 2015 – Jul 2016
  • Utilized ARIMA family models for sales forecasting across various markets for a large cosmetics firm.
  • Created and managed a PostgreSQL database, handling data in formats such as CSV and Excel.
  • Developed automation scripts in Bash and R integrated with CRON jobs to streamline data loading.

Fellow

Data Science Europe · Dublin, Ireland Jan 2015 – Mar 2015
  • Acquired fundamental data science skills covering SQL, Hive, R, and various machine learning models.
  • Analyzed New York City taxi data to predict pickup probabilities by time and location within Manhattan, using Bayesian statistics and Random Forest.

Postdoctoral Researcher

Catalan Institute for Nanoscience and Nanotechnology (CIN2) · Bellaterra, Barcelona, Spain Nov 2014 – Dec 2014

Postdoctoral Researcher

UAB Physics Department · GNM3, Bellaterra, Barcelona, Spain Jul 2013 – Nov 2013

Tech Stack

Languages
Scripting
ML Frameworks
GPU Kernels
Inference & Serving
Profiling
Data
Databases
Visualization
MLOps
CI/CD
Other Tools

Contact

Email: contact[@]agramunt[dot]me

If possible, send me your email encrypted with PGP:

Public Key: email_pk.asc

Figerprint: 1DA7 CE68 83F5 02BB AD56 1048 6E73 C83B 19A5 9D3E

Verify that the key fingerprint is correct with

1
2
 gpg --import email_pk.asc
 gpg --fingerprint paste_my_email_here