Hello, I'm
Rayen Dhahri
I make state-of-the-art AI run where it matters — compressing and deploying vision and language models on constrained hardware through quantization, pruning, and distillation. Machine Learning Research Scientist in Munich, published at NeurIPS, with industry experience across optics, automotive, and semiconductor leaders.
Highlights
What I Do
Model Optimization
Quantization (PTQ/QAT), pruning, sparsification, and distillation of deep networks — from foundation models to production-grade deployments.
LLM & VLM Efficiency
Making large language and vision-language models faster and lighter with low-bit inference, structured sparsity, and kernel-level optimization.
Edge & Accelerator Deployment
NPU-first deployment on Jetson, Hailo, Rockchip, and other accelerators — with hard latency, memory, and energy budgets.
Research & Publications
Published at NeurIPS 2024, EurIPS 2025, and AABI. Reviewer for top ML venues. Active work on efficient AI for the edge.
Interactive
Model Compression, Visualized
A live look at what I do: a dense network is pruned down by weight magnitude until only ~20% of its connections remain — while the signal keeps flowing. Hover over neurons to inspect their connections.
Blog
Latest Posts
Practical notes on efficient AI, model compression, and lessons from deploying ML in production.