nCompass Technologies

nCompass Technologies provides AI agents that identify and resolve CPU-GPU bottlenecks in AI platform workloads, including generative recommender, ranking, and LLM serving systems. The service addresses performance issues across CPU pipelines, serving and scheduling, and custom kernels, delivering changes such as lower end-to-end latency, higher output-token throughput, and increased requests per second on the same GPUs.

The agents work across CPU pipeline optimization, serving and scheduling, and custom kernel development, with integrations for coding agents such as Claude Code and Copilot CLI. The service supports vLLM, SGLang, and custom PyTorch stacks on H100 and B200 hardware. Agents can run as SaaS (SOC 2 Type II), in VPC, or on-premises using a customer's own LLM endpoint. Engagements begin with a scoping call, after which the agents analyze serving code and reference workloads, then deliver pull requests with before/after measurements for the customer's engineers to review and merge. Correctness and bit-wise accuracy checks are applied to proposed changes.