Background
Infrastructure first, then AI on top
AI agents that touch a production network need to understand what they're touching. Two decades of routing, wireless, security and cloud operations taught me where the sharp edges are—and why the same decision that's safe in a lab can break three hundred clients in a venue on a Saturday night.
That background shapes every product I build: how changes are confirmed before they land, how credentials are stored, how errors are surfaced. The AI work is new; the instincts about failure modes are not.
01
Enterprise Infrastructure
20+ years designing, building and operating enterprise networks: campus and data-center routing, wireless architecture, security, Linux, storage, virtualization and cloud. I ran the systems others called me in to fix.
02
Technical Pre-Sales
12+ years as a solutions architect: translating infrastructure complexity into customer outcomes across many verticals. Writing proposals, running POCs, explaining tradeoffs at the whiteboard.
03
Into AI
Completed the Stanford / DeepLearning.AI Machine Learning Specialization and the Hugging Face Agents course, then shipped: a ticket-resolution agent that SSHes into switches, zero-touch AP and switch provisioning through a chat interface, and IoT device fingerprinting. All public on GitHub.
04
NeuralConfig Today
NeuralConfig LLC builds and runs AI products and network tools in production on Cloudflare, and offers technical consulting. NVIDIA Inception member; Cloudflare for Startups; Claude Startups member. Focused on agentic systems that operate across network, cloud and security boundaries.
Selected Work
Three of the eight, up close
Each connects to a real system and handles production data or live calls.
R1-MCP
Claude operates your RUCKUS One network through one connector.
1,572 RUCKUS One REST operations through a remote MCP server, plus full-text search over the 544-page user guide. Built for network admins, MSP engineers and PS teams—interrogating live state, generating as-built packets, running or undoing config changes. Scripts run in an isolated V8 sandbox, 30-second cap, RUCKUS One API only. Write Guard (opt-in): mutating calls previewed and approved before they land. Also: SZ-MCP (1,354 ops) and CP-MCP (175 ops).
StrandCalls
AI voice receptionist: real calls, real bookings.
Answers business calls in natural conversation—no IVR menus. Follows the business's own rules, books appointments into Google Calendar, and routes urgent calls to a designated number. Realtime voice over WebSockets on Cloudflare Durable Objects, Twilio telephony, Stripe subscriptions. NeuralConfig runs its own business line on it: call (910) 994-3824 and talk to it.
NeuralRepo
Capture ideas anywhere. Find them when they matter.
Full-stack AI SaaS for capturing and developing ideas. Capture from CLI (npm package @neuralconfig/nrepo), web, Claude Code over MCP, Siri, email or REST API. Semantic search over vector embeddings, duplicate detection, mind map, bring-your-own-key AI through Cloudflare AI Gateway. Auth, Stripe billing and multi-tenant data on Cloudflare D1. Free tier and Pro plan.
Approach
How I build AI that touches production
These aren't policies on a slide—they're decisions I made while building something that touches production systems. Each one is real and visible in R1-MCP.
- Changes need a human yes.Write Guard is opt-in. When it's on, every mutating API call is previewed—naming the target tenant—and needs explicit approval that expires in 10 minutes, enforced server-side.
- Least privilege is your choice.The user decides which RUCKUS One API client to save. A read-only client makes writes impossible at the API level.
- Generated code runs sandboxed.Claude's scripts run in an isolated V8 sandbox with a hard 30-second cap. The sandbox can reach no network endpoint outside the RUCKUS One API.
- Big changes run as process.A config copy compiles a reviewable runbook: plan, review, arm, run, verify—and every copy can be undone. A runbook replay halts when a verify gate fails.
- State limits precisely; don't oversell.The product page names exactly which data is encrypted at rest—client secrets, venue snapshots, backups—and which is stored but not encrypted. The boundary is named, not implied.
Open Source
15+ public repositories on GitHub
Five selected projects, covering network automation, AI agents and tooling.
All open-source work →
Toolkit
The stack I know best
Networking, cloud, AI and security—the platforms and protocols I've built on and operated.
AI & Development
Python
FastAPI
Claude
MCP
Agentic AI
OpenAI
RAG
Vector Databases
Swift/iOS
Infrastructure & Security
CISSP
Zero Trust
BGP/OSPF/MPLS
Enterprise Wi-Fi
Wireless Architecture
Linux
Virtualization
Cloud Platforms
Cloudflare
AWS
GCP
Azure
Kubernetes
Docker
Automation & Observability
Network Automation
ZTP
CI/CD
Prometheus
Grafana
Credentials
Certifications and coursework
Industry certifications in security and wireless design, plus structured AI training.
Certifications
- CISSP (ISC2)
- CCNP (Cisco)
- CWDP (Certified Wireless Design Professional)
- CWSP (Certified Wireless Security Professional)
- CWAP (Certified Wireless Analysis Professional)
Let's talk.
If the work resonates, reach out. I'm always open to a technical conversation.