AI infrastructure / systems research
I am interested in understanding how and why intelligence arises from systems at scale—from GPU clusters and model weights to large-scale agent collaboration.
I build these systems and measure how they behave: GPU communication, RDMA networks, memory movement, model execution, and agent-to-agent communication.
Selected systems
agent control plane
Deterministic multi-agent engineering for VS Code: issue graphs, MCP control, isolated Codex workers, and parallel execution through bounded workflows.
agent runtime
Server-side multi-agent orchestration with mailbox queues, agenda tasks, groups, resources, and deterministic project state.
cloud runtime
Agent-native cloud runtime with FastAPI, FastMCP, AGENTS.md policy, route intents, JSON contracts, and validation.
01 / blogs
No articles match this search.
Why persistent specialist agents need conversations organized around purpose, with graph execution, shared context, and professional review.
An experiment using domain skills, persistent roles, and collaborative revision to turn a complex PRD into a detailed cloud work plan.
Scaling to 1,000 agents—and making them productive. My 16 lessons: agents keep getting smarter; our job is to help them focus and work reliably. Most importantly, stay humble and take a rest when you can!
A storage-backed runtime pattern for 1,000 persistent agents on OKE.
Parallel issue execution with leases, isolation, observation, and bounded control.
A control-plane architecture for software that contains agents inside its workflow.
Remote PCIe virtualization that preserves native driver behavior across a network.
Four capability levels: orchestration, reliable execution, workflows, and method discovery.
Language, context, memory, and interaction in AI systems.
Persistent memory, multiple agents, generated skills, MCP integration, and a reusable self-evolution loop inside an application.
Agent-native cloud runtime design with policy, route intents, contracts, and validation.
Compute, memory, and fabric design for future AI systems.
Systems-level thinking for AI networking and infrastructure efficiency.
Network architecture for AI-scale communication and resource composition.
Pipeline-driven heterogeneous computing across Arm, GPU, and x86 cloud resources.
Cost-aware workflow placement for heterogeneous pipelines.
Infrastructure notes on scaling behavior, benchmarks, and large AI workload limits.
Benchmarking RDMA and GPU cluster communication for H100 AI workloads.
Human-guided AI engineering and system building.
Guided AI exploration for bus, network, and architecture design.
Human direction, agent execution, and engineering review.
A storage and filesystem experiment shaped by AI-assisted implementation.
AI networking notes on scaling, limits, and architecture tradeoffs.
Benchmarking as a way to reason about infrastructure behavior under workload pressure.
Accelerator fabrics and the changing shape of AI infrastructure.
02 / projects
Treats sparse experts as a managed working set: CPU memory holds the expert bank while routed experts move into bounded GPU slots for fused execution.
Passive offloadActive cacheReproducible DNA and mRNA sequencing workflows that provision stage-specific x86, Arm, or GPU infrastructure and retain task artifacts for inspection.
ArchitectureCost studyGPU-native MoE offload using CPU host memory as the expert-weight bank for constrained GPU memory.
RepoRouter-aware GPU expert slots for local MoE inference under constrained GPU memory.
RepoPatent-submitted scheduling approach for GPU resource management.
Patent-submitted Jupyter kernel architecture for large GPU fabric execution.
RepoDistributed protocol design for AI-native content delivery and agent coordination.
DemoTCP/IP-over-PCIe/CXL and RDMA-style data movement across high-speed interconnect fabrics.
DemoSwitch-chip and multi-CPU board-level architectures for scalable AI and HPC systems.
Multi-rail HPC computing architecture for production rendering workloads.
03 / talks
04 / patents
Two Oracle inventions submitted in January 2026—GPU resource management and distributed interactive-kernel orchestration—remain unpublished.
Links