AI Infrastructure Engineer
Multi-Agent Systems · LLM Inference & Optimization · GPU Systems & Monitoring
Building systems, not demos.
I build infrastructure for agentic and inference workloads: GPU telemetry, model runtime testing, Kubernetes scheduling surfaces, MCP control planes, and long-horizon agent systems.
My current technical direction is an AI Agent Infrastructure Stack:
- Agent runtime: multi-agent orchestration, tool control planes, and human-in-the-loop workflows
- Inference systems: vLLM service testing, model routing, latency tracking, and baseline comparison
- GPU infrastructure: GPU utilization monitoring, sensor parsing, vendor-aware runtime signals, and scheduling inputs
- Technical writing: source-level notes on agent frameworks, AI engineering, and cloud-native systems
A system that connects agent routing decisions with model runtime telemetry and GPU capacity signals.
Core problems I am working on:
- Route simple and complex tasks across different model tiers
- Use latency, throughput, and GPU utilization as scheduling signals
- Expose control-plane APIs for agents, operators, and inference services
- Benchmark dynamic routing against static model selection baselines
This direction combines my existing work across GPU monitoring, Kubernetes operators, MCP servers, vLLM testing, and agent orchestration.
| Layer | Repositories | Signal |
|---|---|---|
| Core Agent System | SwarmMind, deer-flow | Multi-agent systems, long-horizon task orchestration, agent teams |
| Inference Infrastructure | rinference-operator, vllm_test_tool | Kubernetes inference workloads, vLLM lifecycle testing, runtime automation |
| GPU Systems | go-radeontop, gpu_tools, go-sensors-parser, k8s-gpu-hotremove | GPU telemetry, hardware signals, monitoring libraries, scheduling inputs |
| Agent Control Plane | go-sui-mcp, mcp4meme | MCP servers, agent-tool interfaces, external system control |
| Knowledge Output | deerflow-book, kata-container-skill, k8s-operator-skills | Deep technical writing, source-code study, AI/cloud-native education |
Source-level writing on DeerFlow 2.0 and agent system development. This is my strongest public knowledge asset and the clearest proof of technical communication depth.
A Kubernetes operator for deploying and managing AI inference workloads across multiple GPU vendors and inference frameworks.
Automation for repeated vLLM service lifecycle testing: container startup, monitoring, log collection, shutdown, and stability validation.
GPU monitoring library for AI inference systems. Useful for tracking GPU utilization, investigating runtime bottlenecks, and feeding scheduling decisions.
A Go control-plane server that exposes Sui client operations through APIs. Part of my broader work on MCP-style agent control surfaces.
Languages: Go, Python, TypeScript, Rust, C
Infrastructure: Kubernetes, Docker, Linux, GPU runtimes, cloud-native systems
AI Systems: vLLM, MCP, agent orchestration, inference lifecycle automation
Systems Topics: distributed systems, schedulers, observability, performance testing, control planes
I am consolidating these pieces into an ai-infra-playground:
- GPU telemetry collector
- model routing layer
- inference benchmark runner
- agent control-plane API
- comparison dashboard for latency, throughput, and utilization
The goal is to make the system behavior measurable: lower latency, higher throughput, better GPU utilization, and clearer routing decisions.
M.S. Student in Computer Science at Northern Arizona University.
Research and engineering interests:
- AI infrastructure
- GPU systems
- distributed systems
- cloud-native inference platforms
- agent runtime architecture