Christian Smith

Solutions engineer profile: production AI systems, published where anyone can inspect them, and demonstrated to the people who have to trust them.

Remote (US) · open to hybrid in the Washington DC metro · U.S. Citizen, able to obtain a security clearance

Washington, DC · rnvizion.dev · github.com/RNVizion · LinkedIn

I build production AI systems, publish them where anyone can inspect them, and demonstrate them to the people who have to trust them. My color-computation server is published to the official Model Context Protocol registry and listed in awesome-mcp-servers; AIII, my agent identity and authorization reference implementation, ships Sigstore-signed releases, a CycloneDX SBOM, and an OpenSSF Best Practices Baseline submission at 20 of 21 controls with the exception documented rather than hidden. Quality is gated on published thresholds enforced in CI, not on assurances. Four years customer-facing at the technical end: live AR/XR demonstration and tier-1/tier-2 escalation at Meta, enterprise help desk and productivity consulting before that. Nine shipped projects and 5,000+ tests across the portfolio, every project under CI with its quality gates published rather than asserted. Targeting Solutions Engineer, Solutions Architect, and Sales Engineer roles at AI and developer-tools companies.

Languages: Python, JavaScript, SQL, C++, C#, Java, HTML, PowerShell
AI / ML: Claude API (Anthropic), MCP (Model Context Protocol) / FastMCP, LLM agent orchestration, RAG (retrieval-augmented generation), vector search (ChromaDB), embeddings (sentence-transformers), Gradio, Hugging Face Spaces
Security & Identity: OAuth 2.1 resource servers, RFC 9728 protected-resource metadata, holder-of-key verification (RFC 7800/7638), per-tool authorization scopes, threat modeling, network protocols, OS hardening
Testing & DevOps: pytest, pytest-qt, hypothesis (property-based), syrupy (snapshot), coverage.py, CI-gated LLM evaluation suites, GitHub Actions (Linux + Windows CI), DCO enforcement, Sigstore signing, CycloneDX SBOM, pip-audit, bandit SAST
Platforms & Tools: Git, GitHub Codespaces, PyQt6, Google Cloud Platform, Jira, ServiceNow, Google Workspace, Microsoft 365, Slack, Zoom
  • AIII: Artificial Intelligence Identification Initiative · rnvizion.dev/aiii — Apache-2.0 reference implementation for AI agent identity: an L1–L3 identity-and-authorization layer for MCP servers, built on one principle: resolve or refuse, never guess. Holder-of-key verification (RFC 7800/7638), declarative per-tool authorization, 46 tests with property-based eval gates, and cross-platform CI with lint and SAST. Supply chain: DCO enforced in CI, Sigstore-signed release manifests, a CycloneDX SBOM per release, and an OpenSSF Best Practices Baseline submission at 20 of 21 controls, with the one exception documented rather than hidden. Source at github.com/RNVizion/rnv-mcp-identity, with a live demo on Hugging Face Spaces.
  • rnv-color-mcp · github.com/RNVizion/rnv-color-mcp — Color-computation server on the Model Context Protocol (FastMCP), published to the official MCP registry and listed in awesome-mcp-servers. Ten deterministic tools spanning color conversion, harmony generation, mixing, palette management, WCAG contrast (AA/AAA), and perceptual difference (CIEDE2000). Also an OAuth 2.1 resource server: RFC 9728 protected-resource metadata and enforced per-tool scopes over a read and write split, covered by tests green in CI. One shared engine, callable by any LLM client: the model decides intent, the tool owns the exact value. MIT licensed, live on Hugging Face Spaces.
  • Ask the Corpus · huggingface.co/spaces/RNVizion/ask-the-corpus — Live retrieval-augmented (RAG) assistant answering questions over my published writing: sentence-transformers embeddings and ChromaDB vector search, grounded through Claude with source citations, plus rate limiting and per-call cost guardrails. Quality is CI-gated: a 60-case evaluation suite covering retrieval, out-of-corpus refusal, and false refusal, including deliberate traps, runs on every change to the app, the ingestion path, or the corpus, and fails the build if any of four thresholds regresses: retrieval accuracy at or above 85%, out-of-corpus refusal at or above 90%, false-refusal rate at or below 10%, and public-claim accuracy at 100%. Gradio UI, deployed live on Hugging Face Spaces.
  • RNV Publishing Agent · github.com/RNVizion/rnv-publishing-agent — FastMCP server exposing six tools, orchestrated by a Claude agent loop that publishes to a live site: validate the post, commit and push, poll until the URL returns 200, then update the retrieval corpus. The model makes one decision — publish or don’t — and deterministic code executes the chain, stopping at the first failure. The orchestration is code, not a prompt, and the dry run is the default. Refusal behavior is pinned by a test suite; a self-contained demo runs the real chain against throwaway git repositories with no credentials or network, and CI runs both on every push. Recorded walkthrough at rnvizion.dev/demos/rnv-publishing-agent.
MetaSep 2025 – Present
AR/VR Sales & Support Specialist (Best Buy retail program)Annapolis, MD
  • Run live technical demonstrations of the full Meta spatial computing lineup — Ray-Ban Meta smart glasses and the Quest headset family — translating AR, VR, and mixed-reality architecture into terms non-technical buyers can act on.
  • Own tier-1 and tier-2 escalation across hardware setup, mobile app integration, account and identity configuration, firmware updates, and connectivity troubleshooting; resolve at the point of contact rather than handing off.
  • Act as the technical voice in the buying conversation: qualify what the customer is actually trying to do, map it to product capability, and say plainly when the fit isn’t there.
  • Surface recurring failure patterns and product feedback across both lines to inform onboarding improvements and reduce downstream support load.
Hotbox DCSep 2023 – Aug 2025
Manager, Systems & OperationsWashington, DC
  • Built and managed end-to-end retail operations for a regulated startup from launch, owning IT systems, point-of-sale infrastructure, compliance, security, and staff development.
  • Designed and maintained physical and digital security systems, including access control, surveillance, and risk management frameworks.
  • Oversaw regulatory compliance across local, state, and federal regulations, including licensing and age verification protocols.
  • Led staff training and performance management while delivering hands-on technical and customer-facing support.
VitalystApr 2022 – Aug 2023
Productivity Consultant → Managed Help Desk Agent → Content DeveloperRemote
  • Productivity Consultant: Analyzed client workflows, designed change management plans, and led cross-functional initiatives to drive measurable productivity improvements.
  • Managed Help Desk Agent: Delivered tier-1 and tier-2 technical support across Microsoft 365 and Google Workspace, resolving tickets with measurable client satisfaction.
  • Content Developer: Produced technical educational content — written guides, scripts, audio, and video walkthroughs — to help enterprise users adopt productivity tools. Direct precedent for Developer Advocate content work.

Google IT Support Professional Certificate — including System Administration and IT Infrastructure Services, IT Security, Operating Systems, Technical Support Fundamentals, and Computer Networking.

Google Cybersecurity coursework — Foundations of Cybersecurity · Connect and Protect: Networks and Network Security · Play It Safe: Manage Security Risks.

IBM — Artificial Intelligence Fundamentals.

Southern New Hampshire UniversityDec 2019
Bachelor of Science, Game Programming and DevelopmentManchester, NH
  • Minors: Information Technology, Graphic Design