guangli_Liu xuhuan51

I build reliable AI agents, from tool orchestration and evaluation
to high-performance LLM serving.

Runtime Tools Evaluation Serving

_{M.S. in Computer Science @ Zhejiang Normal University · Graduating 2027 · Open to opportunities}

What I Build

Production AI Agents — evidence-driven workflows, tool orchestration, retrieval, structured evaluation, and failure recovery.
Agent Platforms — sandbox runtimes, streaming events, asynchronous jobs, observability, and long-running task reliability.
LLM Systems — vLLM scheduling, KV-cache-aware serving, quantized deployment, speculative decoding, and GPU performance analysis.

System	What I worked on
Argus & Alice	Production AIOps agent and enterprise agent platform: evidence-based RCA, sandbox runtime, streaming orchestration, evaluation, and reliability.
DBOps Enterprise Copilot	A LangGraph Text-to-SQL agent with hybrid schema retrieval, graph-based join planning, SQL verification, and observable Kubernetes delivery.
LLM Serving Stack	Reproducible serving experiments covering vLLM scheduling, KV/prefix cache behavior, Qwen3 deployment, SLO analysis, and GPU topology.
DSpark Serving Benchmark	Speculative-decoding evaluation across models, quantization modes, concurrency levels, and adaptive runtime selection.

Python · C++ · CUDA · LangGraph · FastAPI · vLLM · Qwen3 · Docker · Kubernetes · Prometheus

_{From agent loops to GPU kernels.

Pixel animation from
Hazy Readme Cards,
used under the MIT License.}