Teams at FriendliAI

Design and optimize high-performance GPU kernels (GEMM, attention, routing) for AI inference across NVIDIA and AMD GPUs. Implement CUDA/C++ and low-level assembly code, build reduced-precision/quantized (FP8/FP4) kernels, benchmark cross-vendor performance, contribute to internal GPU libraries, accelerate multi-modal pipelines, and integrate next-generation GPU features into production.

Own quality for FriendliAI's full SaaS stack, including backend microservices, frontend, model deployments, and inference. Build pytest automated suites, Locust performance tests, Playwright end-to-end tests, and design strategies for validating LLM inference and model deployment workflows.

Design, implement, and optimize GPU kernels, kernel compiler, memory planner, and runtime for low-latency generative AI inference. Analyze performance bottlenecks across hardware and software, collaborate with infrastructure teams, and maintain production profiling, benchmarking, and validation tooling while supporting new model architectures and multi-GPU strategies.