Papers
Papers, technical notes and market research that document how Kamirai Labs thinks about private AI infrastructure, not a marketing blog.
A technical note on routing requests across small and large open-weight models based on task complexity, and the trade-offs between latency, cost and accuracy when cascading rather than defaulting to the largest available model.
A short position paper distinguishing data sovereignty, model sovereignty and operational sovereignty, and arguing that private AI infrastructure needs to address all three to be meaningfully 'sovereign' rather than just self-hosted.
An internal technical reference on structuring KV-cache and context memory across GPU, CPU and NVMe tiers when inference is spread across heterogeneous compute rather than a single accelerator class.
A reference framework outlining the access control, auditability and deployment questions organisations in regulated industries typically raise before adopting private AI infrastructure.
A short internal market note summarising early conversations with academic and research organisations in Ireland and the EU on the demand for privately deployed AI infrastructure over external API access.
