We build AI infrastructure and explore how intelligent systems work.

We experiment with model compression, representations, and execution to understand how AI can work with fewer resources. We turn useful findings into tools and use what we learn to inform future model development.

• i.

  Measurements over manifestos

  Concrete numbers, reproducible setups, and code people can actually run. Claims that aren't measurable get cut.

• ii.

  Systems and learning together

  We work where the algorithms meet the hardware. Inference paths, memory layout, and model structure are studied as one thing, not three.

• iii.

  Open notes, when we can

  We share method notes, scripts, and partial results alongside our papers. Some work stays internal while we test and refine it.

• iv.

  Small on purpose

  We stay small so we can focus. Fewer people, fewer meetings, more time on the actual problem.

• Research

  Papers, technical reports, and reproducible experiments on inference, sparsity, and evaluation.

  Read research→

• Writing

  Method notes, post-mortems, and shorter pieces about what we're working on right now.

  Browse writing→

• Software

  Internal tools we use day to day. Some get released; some stay in-house while we harden them.

  Ask about access→