Analog optical computer for AI inference
A compute architecture that combines optics and analog computation to improve efficiency for AI inference and certain optimization problems, alongside model designs that fit the machine.
London
Jannes Gladrow
ML engineer building models and systems for AI, with earlier work across physics, optical computing for drastic FLOPs/W improvements in AI inference, and stochastic thermodynamics.
I work on machine learning systems and model research. My recent work spans post-training, agentic harnesses, large-scale language modeling, fixed-point models, and optical computing for drastic FLOPs/W improvements in AI inference.
Previously I was at Microsoft Research Cambridge and before that I did a PhD in Cambridge on stochastic thermodynamics, optical tweezers, and machine learning.
Selected work
Projects
Implicit language models are RNNs
Work on implicit state-space models that recover non-linear state transitions with more parallelizable training, scaled to reasoning tasks and large language-model pretraining.
Inverse digital holography using conditional generative models
Conditional generative modeling for phase retrieval in digital holography, aimed at finding holograms that produce small laser patterns with high accuracy.
Faster-than-free-diffusion reaction rates
Experimental work on tuned barrier shapes showing regimes where reaction rates can increase rather than decrease with barrier height, against the usual Arrhenius intuition.
Symmetry breaking of transition-path times
Experimental and theoretical work on when fundamental inversion symmetries in first-passage and transition-path dynamics hold, and how they break down on mesoscopic and molecular scales.
Broken detailed balance in living and active systems
A line of work on non-equilibrium dynamics in biological and active-matter systems, including broken detailed balance in living systems and active filament networks.
Papers
Selected publications
Analog optical computing for AI inference
Implicit language models are RNNs: balancing parallelization and expressivity
Inverse digital holography using conditional generative models
Faster-than-free-diffusion reaction rates
Experimental evidence of symmetry breaking of transition-path times
Broken detailed balance and non-equilibrium dynamics in living systems
Career
Experience
ML Engineer
Meta AI, London
Working on post-training, agentic harnesses, AI product experiences, and ML systems for frontier models.
Machine Learning Researcher
Microsoft Research Cambridge
Research engineering on ML systems, infrastructure, and model work.
AI Resident
Microsoft Research Cambridge
Worked on a customer-facing recommendation system based on NLP, alongside physics and computer-vision research projects.
PhD Student
University of Cambridge
PhD in stochastic thermodynamics, optical tweezers, and machine learning, supported by Winton and Marie Skłodowska-Curie scholarships.
R&D Intern
nanoTemper Technologies, Munich
Rapid prototyping of measurement devices for protein characterisation.
Research Intern
Penn State University
Developed LEED correction algorithms in Python.
Elsewhere
Links
GitHub
Code and repositories
Google Scholar
Publications and citations
Professional profile