I study computer networks and telecommunications at INSAT, in Tunis, with a focus on computer vision, applied mathematics, and machine learning. Most of what I work on ends up being about representing structured data — rotations, graphs, geometry — in ways that don't quietly break the assumptions of the models built on top of them. I'm looking for a master's thesis project starting in 2027.
Applied research intern (part-time), Copenhagen, remote
Working on Inter-X, a multimodal model for recognizing human behavior from video, text, and audio. I represent pose orientation as an element of SO(3)/SE(3) rather than as a flattened vector, which avoids the averaging and distortion errors that come from treating rotations as ordinary coordinates. It measurably improved classification accuracy on two internal benchmarks.
Multimodal cattle behavior recognition (MMCows), Tunis
Built a pipeline that combines video and sensor data to classify livestock behavior. The project led to a separate piece of work on reconstructing 3D shapes from multiple 2D views, submitted to ACIVS 2026 and currently under review. A second paper is in preparation.
Undergraduate research student (part-time), Tunis
Ported a small multimodal training pipeline from PyTorch to JAX, running on TPU. Also worked on attention-based methods for tracking multiple people across camera views when they're partially occluded from each other.
R&D machine learning intern, Paris & Tunis
Worked on inference efficiency for LLaMA-7B (KV-cache eviction under a fixed memory budget), on RLHF fine-tuning of a smaller model, and on a multi-agent system for planning.
R&D AI intern (part-time), Zurich, remote
Worked on synthetic data generation, fine-tuned Mistral-7B on chemistry-related reasoning tasks with QLoRA and DPO, and implemented sparse mixture-of-experts routing.
Computer Networks & Telecommunications, Tunis. Specializing in machine learning and computer vision.
Intensive mathematics and physics track, Tunis.
Computer science track, Monastir.