Fabio J. Fehr

Experience

• 

  University of Cape Town

  Jun 2017 - Nov 2020

  Managing and coordination of the large first year statistics course STA1000. This role requires strong leadership and organizational skills to manage the committee and organize the events for the club. It requires a professional relationship with surf brands and related organizations. The club won "Club of the Year " and the "Transformation" award during 2019. Managing emails, administration and taking minutes for the committee while getting involved in organization of events. Statistics Tutor For First year course

  • Head Statistics Tutor

    Feb 2018 - Nov 2020

  • Surf Club Chairman

    Sept 2018 - Sept 2019

  • Surf Club Secretary

    Sept 2016 - Sept 2018

  • Statistics Tutor

    Jun 2017 - Dec 2017
• 

  Eighty20

  Jul 2017 - Aug 2017

  Data scientist: Internship

  I worked in a team on a public interest project that created a visualization of the water crisis in South Africa. Using consumption data and dam levels we managed to create an interactive dashboard that visually displays and informs the user about the current situation. Here you can see the final product:http://water.eighty20.co.za/

• 

  Eighty20

  Jan 2018 - Feb 2018

  Data scientist: Internship

  I worked in SQL writing scripts for large databases on eighty20s data portal. This tested knowledge of data wrangling, manipulation and analysis. Concurrently I worked in R and focused on creating a user friendly interface in shiny to cluster data. All in all my skills were tested and stretched in real work examples.

• 

  DataProphet

  Nov 2018 - Dec 2018

  Machine learning engineer: Internship

  I worked amongst a diverse team of machine learning engineers and data scientists. I helped implement transfer learning using RNN and CNN models for financial timeseries data. The key takeaway skills I learned included: Using a Linux operating system as opposed to Windows and VIM as an editor. Using a git workflow for working in a team. Using shell scripts to run experiments and documentation for reusability. Learning the value of python as opposed to R. Coding in a standard way so that it is easier to review and maintain human readability. Setting up virtual environments for python and their importance in industry. Working and deploying my machine learning models on Google cloud through SSH. Show less

• 

  GetSmarter

  Sept 2019 - Sept 2023

  University of Cape TownProviding help with regards to subject matter for an online Business Analytics course. The skills learned:Tableau, SQL, Python and explaining difficult statistical/machine learning concepts to colleges.London School of EconomicsProviding help with subject matter for an online Machine Learning course. The skills learned: Applying dimension reduction and clustering methods for difficult real world datasets. Interpretations with applications for business. Show less

  • Engagement Tutor: LSE Machine Learning course

    May 2020 - Sept 2023

  • Engagement Tutor: UCT Business Analytics course

    Mar 2020 - Nov 2020

  • Subject Matter Expert: UCT Business Analytics & LSE Machine Learning

    Sept 2019 - May 2020
• 

  Idiap Research Institute

  Feb 2021 - now

  Research Assistant
• 

  EPFL

  Sept 2021 - Jan 2024

  Deep Learning Tutor

  I taught for 2 major course in both Natural Language Processing (NLP) and Machine Learning at EPFL. This entailed teaching machine learning and deep learning at a Masters level and supervising NLP research projects at a PhD level.

• 

  UniDistance Suisse

  Sept 2021 - Dec 2021

  Machine Learning Tutor

  I helped lecture + tutor for the UniDistance Programme : Master in Artificial Intelligence. Courses: fundamentals of Machine learning, Natural language processing.

• 

  Amazon

  Jul 2024 - Dec 2024

  Applied Data Scientist

  Improving retrieval in the retrieval augmented generation (RAG) pipeline with Large Language Models (LLMs) for code generation. I designed and trained a lightweight retrieval model for code that incorporates: the semantics of code with dense embeddings; the file directory hierarchy; and the callgraph structure. This improved retrieval performance by over 50% over the baseline model and outperformed an LLM agent retrieval system with orders of magnitude fewer parameters.