A2Sys

a2sys (a2sys.ai) is developing an agent-native inference operating system positioned at the intersection of AI models, software and hardware, described by the company as "AI, SW, and HW on Memory-centric Infrastructure." Its stated aim is to connect agentic AI workloads to underlying system realities such as execution reliability, inference throughput and GPU cost efficiency.

The scope of work implied by the company's open engineering roles spans an agent platform layer (a backend and runtime that defines and executes agents, with recorded execution state, memory, tool calls and model calls for traceability), an LLM modeling layer (agent fine-tuning plus model compression and optimization aimed at small, specialized models that meet quality targets on minimal GPU resources), a serving layer built on inference engines including vLLM, SGLang and TensorRT-LLM, and an infrastructure layer running on Kubernetes and GPU clusters covering deployment automation, networking, security and observability.

Founding story

a2sys was founded by a team drawn from semiconductor, AI systems research and large-scale AI serving backgrounds. CEO Dongsoo Lee holds a Ph.D. from Purdue University in memory design, VLSI testing and emerging memory, and previously worked as a research staff member at IBM T.J. Watson Research on Power9/10 CPU, NPU design and FP8 training, as principal engineer and AI model compression lead at Samsung Research, and as EVP at NAVER Cloud responsible for AI computing strategy and efficient serving systems. Chief Research Officer Minsoo Rhu is an endowed chair professor at KAIST working on AI computing systems and GPU/NPU hardware-software design, with a Ph.D. from the University of Texas at Austin, prior roles as a research scientist at Meta and senior research scientist at NVIDIA, and service as program chair for MICRO 2025. Both Lee and Rhu are listed as members of Korea's National AI Strategy Committee. Chief Product Officer Baeseong Park previously worked on optimized transformer kernels at Samsung Research and led AI serving systems at NAVER Cloud, and Chief Strategy Officer Se Jung Kwon holds a KAIST Ph.D., was a staff engineer in model compression at Samsung Research, led AI-hardware partnership and strategy at NAVER Cloud, and is an adjunct professor at KAIST with the NAVER-INTEL-KAIST AI Research Center.

Traction

The company is actively recruiting, with open positions for an agent platform engineer, an LLM modeling engineer, an LLM inference engineer and an AI infrastructure engineer, and it operates its own serving stack and GPU clusters.

Latest developments

a2sys is hiring across agent platform, LLM modeling, LLM inference and AI infrastructure roles, and invites inquiries on partnerships and technical collaboration.