AI Systems Architect · Building the tooling layer between language models and real-world infrastructure
I build infrastructure that connects AI models to real systems — giving agents the ability to query, deploy, orchestrate, and take action across live environments through structured tool interfaces.
My focus areas:
-
MCP Server Architecture — Designed and built 7 production MCP servers exposing 200+ tools. Published BMCPS v3.0, an open standard for consistent, safe tool interfaces. Built a meta-server that generates new MCP servers automatically from the standard.
-
Multi-Agent Coordination — Built a 6-agent orchestration system with 33 team templates for parallel task execution. Model-agnostic — works with Claude, GPT, Ollama, and local models.
-
Semantic Memory for Agents — Created a persistent memory system using PostgreSQL + pgvector that gives Claude long-term recall across sessions, with a browser extension and dashboard.
-
Technical Writing — 45 posts at operationalsemantics.dev documenting the full arc from experimental MCP development to published standard.
| Project | What It Does |
|---|---|
| longterm-memory-macos | Semantic memory for Claude — PostgreSQL + pgvector with browser extension and dashboard |
| ai-team-orchestrator | Model-agnostic multi-agent coordination — 6-agent tmux workspace, 33 team templates, 19 MCP tools (deep dive) |
| mcp-factory | Meta-MCP server — auto-generates standard-compliant MCP servers from a single schema (how it works) |
| evm-chains-mcp-server | 111 tools across 7 networks — demonstrates MCP at scale for complex multi-system orchestration |
| ccxt-mcp-server | Unified MCP interface to 106+ financial data APIs with real-time arbitrage detection |
| academic-lectures | Lecture materials from teaching engagements at Ivy League institutions and Oxford |
Guest lectures at Harvard, MIT, Princeton, Cornell Tech, NYU, and Oxford. Keynote at Web3 Summit Amsterdam. Topics include agentic workflows, LLM architectures, AI-driven automation, and building developer tools for emerging platforms.
See the full lecture portfolio: academic-lectures
- Natural-language control surfaces for real infrastructure
- Multi-agent cooperation and task decomposition
- Tool-call runtimes and safe execution patterns
- Autonomous workflow design with fail-safe operational semantics
- Model behavior at the boundary of reasoning and action
Languages: TypeScript, Python, Rust AI Platforms: Claude, MCP, Ollama, LM Studio, OpenAI Infrastructure: PostgreSQL, pgvector, Cloudflare Workers, Node.js, bun
Blog · BMCPS Standard · Portfolio · LinkedIn · X