Luís Freire

I am a Computer Science graduate with 3+ years of experience as a Software Engineer.

Went back to study the Masters to deepen my knowledge in AI, but I mostly see myself as a problem solver so I am open to different roles and opportunities.

Education
M.Sc.Eng. Computer Science and Engineering

Technical University of Denmark (DTU)

Sep 2023 - Aug 2025

B.Eng. Software Engineering

VIA University College

Aug 2018 - Feb 2022

Courses
Elements of AI - Introduction & Building AI

University of Helsinki

Feb - March 2021

Traits

  • Eager to learn
  • Team player
  • Creative thinker
  • Brings the good mood

Languages

  • Portuguese
  • English
  • Spanish
  • Danish
  • French
  • Native
  • Proficient
  • Fluent
  • Interm.
  • Beginner
Work Experience

Student ML Engineer

the LEGO Group

Feb 2025 - Aug 2025

6 months

  • - Conducted research on customizing Large Language Models (LLMs) for domain-specific language (DSL) code generation
  • - Implemented and evaluated retrieval-augmented generation (RAG) using vector databases and parameter-efficient fine-tuning (LoRA/PEFT) to fine-tune model on DSL coding tasks (PyTorch, HuggingFace)
  • - Developed and leveraged synthetic datasets with foundation models for DSL used for both fine-tuning and as context
  • - Created a pipelines for synthetic generation, fine-tuning, evaluation and deployment of the models (Spark, Databricks, Mlflow)

Associate Software Engineer

the LEGO Group

Feb 2022 - Jan 2025

3 years

  • - Worked as a full-time engineer for about 1,5 years in Agile environment and transitioned to part-time within the same role after starting the masters program
  • - Improved my knowledge on handling large amounts of data whilst working on a RESTful application in ABAP Cloud (SQL, ABAP)
  • - Made the difference by proactively developing new CI/CD pipelines reducing deployment time by 90% and presented the findings to the whole department as it benefited other teams (Github Actions)
  • - Contributed to internal community driven packages to extend compatibility to teams using ABAP BTP(Node.js, CAP, Swagger)
  • - Proactively developed a TS library that centralized common JS code in our 100+ apps improving testability, reliability and code maintainability (SAPUI5, Typescript)

Back-end Developer Intern

Swap Language ApS

Sep 2020 - Feb 2021

6 months

  • - Worked mostly within the .NET Core environment focusing on backend but also gained experience in frontend technologies (C#, JS, Razor)
  • - Developed my understanding in cloud computing (Azure, Kubernetes, Docker)
  • - Designed and developed two microservices, one for Data Analysis and another, for sending Emails (MSSQL, CosmosDb)
  • - Learned more about QA practices, focusing on integration and unit testing
Other relevant experience

Master Thesis Project

Arxiv Paper

Researched customization of small LLMs (~1B parameters) for code generation in Python, Scikit-learn, and OpenCV using synthetic datasets generated by larger models. Evaluated Few-Shot Learning, RAG, and LoRA-based PEFT fine-tuning for improving small LLM performance. Showed that synthetic data effectively achieved the highest accuracy in code benchmarks through fine-tuning, while few-shot and RAG provided cost-effective improvements.

OrangeBush: Houseplant Manager

  • - Houseplant database that allows for image recognition of the user's plants based on image data collected by the members of the team
  • - Developed the backend deployed on Azure as a REST API (C# .NET, MSSQL)
  • - Fine-tuned a pre-trained model on our augmented dataset for image classification later deployed as an Azure Function
  • - Developed the CI/CD for testing the application before deployment (Github Workflows)
  • - The website was developed as a static web application (Vue.js)
Outreach

Re:wair

Ongoing

Volunteer

Currently helping in this NGO that fights fast fashion and clothing waste by providing youth knowledge, tools and forming a community around this cause.

ValHacks

Jun 2024

1-day ML Hackathon

The goal was to compress large audio files and classify unlabeled sounds (e.g., babies snoring, crying) into predefined labels. We explored Active Learning and pre-trained models, but faced data pre-processing issues and while some teams achieved good audio compression, no working model was ultimately developed given the time constraints.