Parking traces processed in recent generative data workflows.
AI engineer working across research and production systems
Building machine learning systems for real-world use.
I work across autonomous driving, deep learning optimization, and MLOps infrastructure. At Valeo, that means shipping pipelines, accelerating JAX workloads, and helping models move from experiment to fleet-scale execution.
Observed MFU range on one recent generative pipeline rather than a theoretical peak.
Reduction in one internal height-map processing path.
Long-context target for differentiable Sinkhorn attention work.
Focus Areas
Where I tend to do my best work
The through-line is simple: I like technically hard systems where algorithmic quality and engineering quality both matter.
Shipping models into constrained environments
I care about the ugly middle between notebooks and deployment: dataset movement, observability, inference constraints, and the handoff between research code and real systems.
Turning theory into fast, inspectable implementations
My projects often start from papers or first principles and end in working kernels, clean evaluation loops, or experimental tooling that can actually be trusted.
Designing pipelines with fewer hidden failure modes
I like replacing fragile manual steps with explicit orchestration, better test coverage, and systems that recover well when data, dependencies, or hardware are imperfect.
Selected Work
Recent systems and research threads
Generative data and infrastructure for autonomous systems
Built and optimized internal ML workflows spanning generative augmentation, trace processing, orchestration, and cross-cluster scheduling for perception workloads.
Block-wise differentiable Sinkhorn attention
Designed a memory-efficient optimal transport kernel for long-context alignment with a custom backward pass and TPU-oriented JAX/Pallas implementation.
Developer tooling and mocking for faster preview loops
Built state and API mocking systems, configuration abstractions, and LLM evaluation work to reduce friction in production and internal engineering workflows.
Books
What I am reading and what has stayed with me.
Currently Reading
A few of the books I am in the middle of at the moment.
Loading current reads
Drawn from my public Goodreads shelves.
Writing
I also write about coordination problems, AI, and incentives.
The essays are less about recapping the news and more about building better models for how technology, institutions, and people behave under pressure.
Tragedy of the Correlated Commons
On why prestige optimization can flatten out weird, frontier-seeking talent.
The Communist Constraint
Using gradient descent to think about planning, markets, and adaptation.
Compounding Genetic Capital
An essay on selection, compounding effects, and long-horizon divergence.