Alessandro Fontanella
Research Engineer @Snap Inc.
Signup · Get unlimited contacts
WORK HISTORY
Research Engineer @Snap Inc.
London, GB
Owned the pose-controllable video generation pipeline. Leveraged PyTorch DDP/FSDP on multi-node clusters to train the 30B+ parameter diffusion transformer, achieving a human quality rating of 4.37/5.0- Developed and benchmarked conditioning architectures and attention mechanisms to improve identity preservation and cross-frame visual consistency, achieving 21% lower average LPIPS than baseline models- Optimized inference speed for pose-controlled video generation by leveraging distilled architectures and custom LoRA adapters, reducing inference latency to 15 seconds for 61 frames- Built an evaluation framework combining automated metrics (LPIPS, FVD, SSIM, CLIP score) with large-scale human studies to measure generative quality, consistency, and controllability- Developed causal temporal-consistency training and lightweight memory mechanisms for real-time image-to-image AR effects, improving frame-to-frame stability under live-streaming latency constraints and achieving 35% lower FVD and 22% lower temporal LPIPS.
EDUCATION
Università degli Studi di Torino
Master of Science - MS, Mathematics
The University of Edinburgh
PhD in Machine Learning
The University of Edinburgh
MScR in Artificial Intelligence
ABOUT ALESSANDRO FONTANELLA
Machine Learning Research Engineer with a PhD specializing in generative modeling for images and video. Expertise in designing and training large-scale diffusion models for controllable generation. Industry experience at Snap and Huawei, with publications at NeurIPS and ICML.
This profile is compiled from publicly available professional sources. Unifers is not affiliated with or endorsed by LinkedIn. Request removal of this profile.