Summary
Forward Deployed Engineer / Applied AI Engineer with 9+ years of professional experience (U.S. Army team leadership, customer-facing technical sales, operations) and two years designing, shipping, and operating production LLM agent systems end to end: an autonomous outbound agent with defense-in-depth guardrails and cryptographically signed, independently verifiable action receipts (zero unaudited sends); human-in-the-loop approval gates (folders a human moves, enforced in CI) across a multi-project fleet; and production web platforms in TypeScript/Next.js on Postgres and Vercel with CI/CD. Python-first and self-taught; I turn "be careful" into machinery, verify instead of asserting, and ship into the environment the customer actually runs in. Claims are backed by live sites, a public proof artifact with its verify command, and demos on request. Targeting Forward Deployed Engineer, Solutions Engineer (AI), and Applied AI Engineer roles.
Technical Skills
- Languages and Frameworks: Python (stdlib-first production code); TypeScript; Next.js (App Router); React; SQL (Postgres); PowerShell; HTML/CSS
- AI and LLM Systems: LLM agent architecture; tool calling / function calling; MCP (Model Context Protocol) servers; multi-agent orchestration; human-in-the-loop approval gates; guardrails and defense-in-depth (deterministic, input-independent gates; withhold-only reviewer authority); prompt-injection defense and red teaming; evaluation frameworks and calibration design; structured outputs (schema-constrained agent returns); retrieval (multi-hop recall over a typed knowledge graph); agent observability, monitoring, and alerting; Claude and frontier-model APIs
- Security, Trust, and Auditability: cryptographic verification (Ed25519 signatures, HMAC-SHA256, SHA-256 hash chains, Merkle consistency proofs per RFC 6962); secrets management (DPAPI-encrypted secrets, environment-scoped secrets, fail-closed when absent); auditability (append-only ledgers, signed attestation receipts, tamper-evident stamps); AI governance (machine-verified AI-disclosure invariants, tiered autonomy); dependency audits, security headers, and per-IP throttling; timing-safe comparisons
- Platform and Delivery: CI/CD (GitHub Actions, Vercel); Git; REST API integration (Zoho CRM, Notion, Cal.com, Hunter.io, Upstash); cloud services (Vercel, Cloudflare DNS and R2, Neon Postgres, Upstash KV); database migrations tested up and down in CI; Windows automation (PowerShell, scheduled tasks); local-first / privacy-first architecture
- Customer-Facing and Leadership: customer-facing consultative technical sales; stakeholder communication; team leadership and training; technical writing (playbooks, handoffs); UX and brand design (A.A.S., Graphic Design)
Professional Experience
Co-founder & Founding Engineer - membrAIn (AI security venture, stealth)
Charlotte, NC | 2024 - Present
- Built and operate a production autonomous outbound email/CRM agent integrated with Zoho CRM, Cal.com, and Hunter.io through their APIs: a staged pipeline (prospect, qualify, personalize, adversarial review in fresh context, gated send) with defense-in-depth guardrails - deterministic, input-independent send gates and a model reviewer holding withhold-only authority, so a prompt-injected model cannot authorize a send. Zero unaudited sends.
- Designed an Ed25519-signed attestation-receipt chain for every governed agent action (send / handoff / block): append-only, vendor-neutral, independently verifiable, one attestation format shared by agent and product gateway.
- Hardened the release path: a mock-mailbox regression sandbox exercises the full respond loop before any live credential is touched; DPAPI-encrypted secrets management; deliverability-gated enrichment enforcing a <2% bounce policy.
- Implemented machine-checked AI governance in production: AI-disclosure rules as machine-verified vault invariants, tiered autonomy with an auditable growth ledger, and automated claims-drift monitoring against a declared source-of-truth order.
Independent AI Systems Engineer - multi-project agentic fleet
Charlotte, NC | 2024 - Present
- Shipped and operate brewtooth.org, a production Next.js/TypeScript site on Postgres (Neon), Cloudflare R2, and Vercel: 155 shop profiles, 42 neighborhood pages, and 17 tracked database migrations, each verified by name in the database catalog; 526 tests plus lint, typecheck, build, and a dependency audit, all fail-closed in CI/CD (GitHub Actions); six security headers on every route and per-IP login throttling; deploys verified by a build-marker health endpoint; the single model-spend path is budget-capped and produces drafts a human approves.
- Built the Charlotte Table Tennis Club ecosystem (cttconline.us): five workspaces in one repository, decomposed into 36 stage contracts on eleven shared packages; human-in-the-loop approval gates implemented as folders (pending/ to approved/) with deliberately no approve() API, enforced repo-wide in CI; ten mutation-gauntlet scripts in CI prove the tests would catch a real break; migrations run up, down, and up again on real Postgres in CI.
- Built Rhyza, a verification-ranked knowledge graph for agents: a localhost API (87 routes), a 16-tool MCP (Model Context Protocol) server so any agent can query the graph through tool calling, Merkle-anchored receipts, valid-time (as-of) queries, and multi-hop retrieval that fuses keyword search with rank blending - a chain takes the label of its weakest link, so guesses can never present as verified.
- Authored Goshuin, a tamper-evident, self-verifying document engine used as the system of record across five workspaces (hash-chained sections, machine-checked invariants, HMAC-signed stamps, an embedded selftest; Merkle consistency proofs per RFC 6962 as a companion pattern) - 600+ ledgered, verifiable changes.
- Operate the Dojo, a fleet-wide commons every project subscribes to, where written rules become enforced machinery: standing rules kept in a hash-chained vault and cited by ID; a deterministic, injection-immune intake pipeline; a daily patrol with failure alerting; a read-only headless distillation session with no write tool, so hostile text in a harvested git log has nothing to act on; human ratification of every rule; and pre-execution hooks (PreToolUse) that refuse a tool call before it commits. Its append-only engineering failure ledger (290+ distilled, provenance-tagged prohibitions) is enforced across all projects.
- Designed ground-truth-first evaluation frameworks and calibration protocols for probabilistic systems (pre-registered resolution windows, uncertainty-gated promotion, replayable event-stream instruments, benchmark CI with frozen anchors) and an adaptive training-curriculum engine (Bayesian knowledge tracing, Glicko-2 uncertainty gating), plus a branded product design system (explainers, UX).
- Run multi-agent orchestration as reusable workflow scripts: a tiered build cadence (planner, parallel builder agents, adversarial reviewer agents returning structured, schema-constrained findings with must-fix flags, and a mechanical-edit lane) with per-run token metering, applied across the fleet.
- Armed Security Officer, Charlotte, NC | 2025 - 2026 (concurrent with the venture work above)
- Chef (Kitchen Lead), Noble Smoke, Charlotte, NC | 2023 - 2024. Led a kitchen team in high-volume operations.
- Solar Sales Associate, Blue Raven Solar, Charlotte, NC | 2021 - 2022. Customer-facing consultative technical sales; exceeded quota.
- Sales and Operations, Archer Paper Co, Charlotte, NC | 2019 - 2021. Customer-facing sales and operations.
- Airborne Infantryman, U.S. Army, Fort Bragg, NC | 2013 - 2017. OEF XIV deployment (Afghanistan); led a 4-soldier team; accountable for $2M+ in equipment; honorable discharge.
- Design Associate / Regional Warehouse Manager, Abercrombie & Fitch Distribution Center, New Albany, OH | 2011 - 2013
- English Tutor (independent), Sagamihara, Japan | 2009 - 2011
Education
A.A.S., Advertising and Graphic Design - Central Piedmont Community College, Charlotte, NC | 2022 | GPA 3.8 | Phi Theta Kappa Honor Society