Kam Basra — UK. Inference/serving engineering; first-principles debugging of LLM inference systems. Currently contributing to vLLM around KV cache, scheduling, and correctness.

Recent: #39146 — replication & discrimination study of temperature-0 nondeterminism under concurrency (batch numerics, not KV corruption; raw data published) · #52747 — docs, cache-usage reporting & prefix-cache retention · #52771 — root cause + fix, OffloadingConnector under MTP/EAGLE (hardware-validated by the reporter).

Open to reviewing in the KV-cache/scheduler area.