Douaa
Kdidar
I rebuild the fundamental version of things before trusting the library — a transformer trained from scratch, a Monte Carlo Tree Search opponent, a smart-contract ticketing system — then measure exactly how well each one holds up.
cumulative average
Baccalaureate ranking
Hackathon, 2026
Precision, then scale.
Douaa is a second-year Computer Science student at EPFL, currently holding a 5.94/6.00 cumulative average. Before EPFL, she ranked 6th nationally out of more than 300,000 students in the Moroccan Baccalaureate and graduated as her high school's valedictorian across every major, out of 200+ students.
Her projects follow one habit: rebuild the core mechanism from scratch before trusting it. That's meant training a transformer language model on campaign-rally transcripts, writing a parallelized Monte Carlo Tree Search agent to play the board game Azul, and hand-implementing decision trees and backpropagation during a graduate-level machine learning specialization.
The same instinct shows up off-screen — she has soldered the pressure sensors on a student rocket, and organizes sponsor meetings and cross-team coordination for EPFL's Rocket Team.
- Co-organized by Morocco's Space Industry initiative, the U.S. Embassy in Morocco, and UM6P
- Selected as one of 30 high-school participants nationally, from over 1,600 applicants
- Built a rocket end-to-end — programmed and soldered onboard electronics, including speed and pressure sensors
- Studied aerodynamics, rocket design, numerical methods, and aviation history, taught by engineers and researchers from NASA, Blue Origin, the University of New Hampshire, and Google DeepMind
- Organized internal and external events, including presentations to professors, sponsor meetings, and coordination across subteams
Full from-scratch implementation of the Portuguese tile-laying game Azul — complete rules engine plus a custom AI opponent running parallelized Monte Carlo Tree Search across large-scale self-play simulations. Built as EPFL's CS108 Advanced OOP semester project.
Autoregressive transformer built from scratch and trained on 2016 presidential rally transcripts, extending Andrej Karpathy's "GPT from Scratch" into an original project covering attention, positional encodings, and tokenization.
Smart-contract architecture for a blockchain-based ticketing platform, built at the SUI Blockchain Hackathon to shut out predatory ticket reselling and protect the end-user experience at scale.
Contributed to the SHAP library, extending unit-test coverage across several helper classes in the Shapley-value model-explanation pipeline.
Coursework from Andrew Ng's Machine Learning Specialization and Stanford CS229 — GLMs, SVMs, kernel methods, PCA/ICA — with decision trees, backpropagation, and a fully configurable neural-network trainer implemented from first principles.
Implementation of the game strategy used to place 2nd at The ConquerHack Hackathon, 2026.
Languages
Libraries
Theory & Algorithms
Let's build something precise.
Open to internships, research collaborations, and hackathon teams. Reach out directly, or find the work in progress on GitHub and GitLab.