Resume

Built and operate a set of systems that let an AI agent run real infrastructure safely, and shipped health-data tooling that I depend on daily.

• MCP platform: designed and shipped 6 Model Context Protocol servers (container control, Postgres and observability, knowledge vault, Discord, Stitch, read-only work data) that let Claude operate a live homelab, with none on the public internet: LAN-only binds or local stdio, bearer-token auth, a socket proxy and a 17-container allowlist, credential redaction and an approval gate on anything that changes state.
• Agent process engineering: authored 4 Claude Code subagents (two read-only investigators, an operator that makes only approved changes through a restricted no-shell key, an unanchored code reviewer) and 9 skills that encode an operating method (pre-push secret scanning by live credential value with self-tested patterns, knowledge-base writing and drift audits, firmware deploy, QA testing, maintainer-standard code review) so an AI session follows the process without being reminded. Security-scanned 13 third-party skills before installing them.
• Health-data platform: built and run a self-hosted insulin and glucose service that ingests Dexcom CGM readings every minute and Glooko pump data, deduplicates across writers, calculates doses and insulin on board, and serves a dashboard, a Garmin watch widget, and three LED clocks.
• Embedded and devices: got 12 pull requests merged into the open-source nightscout-clock firmware (alert windows and sounds, a daily face schedule, a rebuilt settings page, settings backup), with 12 more in review and ported the concept to a second platform on embedded Linux through reverse engineering.
• LLM-assisted operations: built an always-on Discord error bot with local-LLM (Ollama) root-cause analysis, auto-heal restarts, AI digests and a conversational interface, ran it, then retired the LLM parts in June 2026 for a deterministic n8n pipeline that groups Docker log errors into incidents. I review that weekly with a Claude Code skill, and every row carries a human decision, now 511 labelled rows that form the dataset for a planned eval harness.
• Security discipline: ran a full security scan and architecture review of the lab, moved hosts to key-only SSH, rotated credentials with proof-by-refusal, and wrote the drift checks that detect their own death.
• Knowledge system: maintain a validated markdown vault of runbooks, journals and gotchas, served to agents over MCP and audited for drift.

Stack: Python · FastMCP · Postgres · Docker · n8n · Cloudflare · C++ · ESP32 · Monkey C · Claude Code

City-scale public safety program: real-time crime center platform and its mobile apps for a major U.S. city.

• Lead the dev and QA team for a six-plus app mobile suite (311 dispatch, digital Activity Log, alerting and subscriptions, search, 911 jobs) supporting 700K+ weekly logins across mobile and desktop.
• Rebuilt 311 Mobile into a full operational workflow: supervisor and self-assignment, multi-job handling, arrival and closure tracking, spam detection and grouping, citywide case-management integration. Now handling 30K+ requests and 300K+ job views weekly.
• Led the pilot design and measurement framework for the new 311 workflow: validated 580 mobile-managed requests against 1,740 legacy cases, cutting average resolution time from 1.36 days to 0.26 days (80%+ faster).
• Replaced the paper memo book with a mandatory digital Activity Log for every patrol officer, relied on by supervisors, internal affairs, courts, and investigators. Now 1M+ entries and 100K+ users weekly.
• Own real-time alerting and subscription-based intelligence distribution: drones, plate readers, persons and vehicles of interest, geofences, and cross-agency alert sharing. Delivered the drone-as-first-responder integration with live video, deployment tracking, and metrics.
• Introduced AI-assisted analysis: built an MCP server giving Claude read-only SQL access to service-request data for plain-language querying.

Same public safety program, earlier phase.

• Owned a real-time crime center and counter-terrorism platform used by 30,000 officers (20,000 daily active), fusing a citywide sensor network of CCTV feeds, license plate readers, 911 and records data into one operational picture. Credited with an estimated $50M per year in operational savings.
• Owned desktop, mobile, and tablet apps delivering live alerts, records, and camera access.
• Led a multi-million-dollar regional integration unifying multiple source systems.
• Owned requirements as the system of record: structure, versioning, and traceability from requirement to backlog item, test case, and release note ahead of CAB and executive sign-off.

Consultancy; client engagement with BAE Systems.

• Product owner for a tablet and personal-device inflight entertainment platform for BAE Systems.
• Authored FRDs, user stories, and acceptance criteria with UX/UI.

Wealth Management.

• Wealth Management onboarding modernization: cut documentation 50% and pre-populated 50%+ of fields.