Enter my personal space ↗ · Browse repositories
Agent Runtime & Architecture · Inference Infrastructure
Focused on agent execution lifecycles, tool orchestration, and the resource constraints of LLM inference.
- TensorRT-LLM #21470: A 16-GPU KV cache failure, reduced to 4 MiB — Single-GPU reproduction of a native KV cache allocator failure on an RTX 5080, with analysis of capacity limits and request admission.
- Agent systems: execution lifecycle, state persistence, and human-in-the-loop workflows.
- Protocols & integration: MCP, REST APIs, and event streaming.
- Developer infrastructure: observability, automated testing, and workflow automation.
Clear interfaces, explicit state management, and reproducible validation.
Python · LangGraph · Deep Agents · SSE · REST APIs · Git

