Senior Systems Solution Engineer and hands-on applied AI systems builder — almost 20 years in enterprise infrastructure and customer-facing solution engineering, now focused on self-directed agentic AI engineering: multi-agent orchestration, LLMOps, local LLM inference, model evaluation, and memory/retrieval architecture.
I build agentic AI systems from scratch and run them in production for my own work — actively adding features, fixes, and optimizations — and I contribute upstream to popular open-source AI projects.
- CMC OS — a local-first agentic workflow control plane that turns vague goals into structured design, Pipeline/Kanban work items, agent + tool execution, and test/audit gates, with human escalation only at risk or decision points. Running in production for my own workflows, under active development; companion Android app in design. (Public architecture notes + Mermaid diagrams.)
- Hermes Agent — a 7-role multi-agent system (generalist, coder, auditor, orchestrator, ops, pr-scout, researcher) with a Mixture-of-Agents "LLM council" + judge for high-stakes decisions, over durable memory (append-only JSONL + SQLite FTS5 + RRF hybrid retrieval). In active use and continuously improved. (Public architecture notes + Mermaid diagrams.)
- Odysseus Vault Mode — a downstream privacy/isolation feature for sensitive AI sessions (local-model-only execution, isolated from normal memory/RAG/agentic context), structured as an upstream feature proposal.
cmc-os-architecture— public architecture notes for CMC OS: multi-agent orchestration, MCP/tool execution, audit gates, and human-in-the-loop AI workflows.hermes-agent-architecture— public architecture notes for my Hermes-based multi-agent system: role-specialized agents, LLM council/judge orchestration, durable memory, hybrid retrieval, and local AI operations.local-llm-lab— practical local LLM lab: inference runtimes, AI agents, LLM evaluation, memory/retrieval, and GPU infrastructure.odysseus— fork of a popular self-hosted AI workspace (80k+ stars), used for upstream contribution work and agent-workflow exploration.enterprise-ai-architecture-checklists— practical enterprise GenAI architecture, readiness, governance, security, and LLMOps checklists.
22 public PRs (21 merged) hardening two popular open-source AI platforms — tests, integrity checks, safe deployment, and upstream-friendly documentation.
- Oprel (local-AI platform) — 15 merged PRs: binary provenance, SHA256/size integrity checks, safe download cleanup, verified manifest entries, deployment guides, CI smoke/link checks.
- Odysseus (self-hosted AI workspace, 80k+ stars) — 7 PRs (6 merged, 1 open): memory owner-isolation tests, webhook test isolation, hardware-fit CPU fallback, CORS/proxy docs, server metadata preservation, provider detection.
- Agentic AI: multi-agent orchestration, LLM councils / Mixture-of-Agents, tool calling, MCP
- LLMOps & inference: local/cloud LLMs, llama.cpp/GGUF, quantization, KV-cache, long-context, cost/latency/reliability trade-offs
- Memory & retrieval: SQLite FTS5, semantic + hybrid retrieval, RRF ranking, memory architecture
- AI infrastructure: GPU/CPU workload placement, model-role separation, model evaluation for real engineering tasks
- Enterprise foundation: ~20 years infrastructure + customer-facing solution engineering