independent inference research

Coconut Labs

Schedulers, systems notes, and reproducible measurements for shared inference.

Coconut Labs works on the shared layer of inference: scheduling, fairness, cache pressure, and the measurements that keep claims honest.

The lab is small by design. Fewer abstractions between the benchmark, the note, and the code.

The quiet tenant should still have a name.

Recent research

2026-04-19 · research note

A KVWarden Gate 2 note: 53.9 ms quiet TTFT, 61.5 ms under flooder pressure, and 26x better tail behavior than FIFO.

1.14x of solo, 26x better than FIFO

Tenant fairness on shared inference. A quiet tenant stays visible when a flooder arrives.

Apple Silicon scheduler experiments

A local inference thread focused on load, correctness, and tenants that do not politely wait their turn.

people

Coconut Labs is Shrey Patel and Jay Patel, building public research around inference scheduling, fairness, and systems measurement.

Co-founder · Engineer

Engineer and writer. Builds inference middleware between LLMs and the GPUs they run on. Currently building Coconut Labs.

Co-founder · Engineer

Engineer focused on inference reliability and tenant fairness on shared hardware. Co-founder of Coconut Labs.

Building something at this layer? Write us.

updated 9h ago14 commits this week3 repos tracked9 RFC open