FlashKernel
Custom CUDA C++ and Triton kernels for transformer inference — tiled FlashAttention, fused GeLU, RoPE, paged KV-cache — benchmarked with Nsight Compute on T4.
View details →
Abdeljalil Jliouat Applied AI Scientist · Paris
Applied AI & systems engineering
AI systems, from research to production.
I build efficient, reliable AI systems — from GPU kernels to robotic learning. Here I share the code, experiments and engineering decisions behind my work.
What I work on
Three connected areas of applied research and engineering.
Compute & inference
Efficient models, CUDA kernels and the systems that run them.
Robotics & neural signals
Learning from language, interaction and brain signals.
Optimization & energy
Classical and quantum methods, tested against practical constraints.
Selected projects
Explore the architecture, experiments and source code behind four systems.
RoboLLM
Language-grounded robotic manipulation — VLM planner decomposes instructions into sub-tasks, RL policies execute each step in MuJoCo simulation.
View details →NeuroLLM
Foundation model for neural signal decoding — pre-train a transformer on large-scale EEG, fine-tune for motor imagery BCI with frequency-band attention.
View details →QuantumGrid
Quantum-classical hybrid optimization for energy grids — QAOA and VQE applied to unit commitment on real ENTSO-E data, benchmarked against MILP solvers.
View details →Experience
Leading applied ML and AI systems in startups, NGOs, and banking.
-
Founder @AKIOUD AI, Paris
-
Lead Data Scientist, Merci Jack, Paris
-
Lead Data Scientist, Société Générale, Paris
Contact
Available for collaborations, advisory work, and technical leadership roles.