I ship systems in places software is not supposed to work. A Sherpa village in Nepal. A hospital ward in Dublin. A sealift crate bound for the High Arctic. A closed three-mile street in San Francisco.
10 roles since age 14 2 patent filings pending 315 tests on real radios 2 San Francisco Chronicle features
2026 – nowClearText
Technical co-founder · New York
Sole inventor on two patent-pending filings for recovering latent structure from identifiers assumed to be random, using public observation alone. I run the research the way I run everything: kill criteria written before the build, claims that have to be falsifiable, and one product line already killed on its own gates.
2 patent filings pending · sole inventor
Aqqut2025 – now
CTO · edge AI for the High Arctic
I run the software side end to end: an industrial gateway whose store-and-forward egress survives satellite outages and power cycles at a site nobody can walk over to reboot. The hardware rides a once-a-year sealift, so the software has to be right the first time.
778 tests · CI on both repos · staged after the sealift
Dec 2025 – Mar 2026
Frontend developer · now live at wrened.com
I built the frontend for a platform of classroom math activities and the teaching moves behind them: React and Vite, live KaTeX math rendering, a markdown activity pipeline, PDF export of generated activities. I prototyped the original activity engine before the production rewrite.
launched publicly in 2026
The San Francisco Chronicle covered it twice. I organized it: a closed three-mile route from Christmas Tree Point through Noe Valley into the Castro, staffed like a deployment.
San Francisco Chronicle
The preview. After years of injuries and confrontations at Dolores, the Chronicle reported that "a group of skaters will hold a separate event Saturday on Twin Peaks," and walked through the plan I brought: permits, a safety plan the city signed off on, spotters with radios, medics on standby.
Read the article ↗San Francisco Chronicle
The day of. The paper described "a gathering that avoided the injuries and police confrontations of the past," named me as the organizer who conceived it, and quoted the speech I gave from Christmas Tree Point before the first drop.
Read the article ↗"We're definitely trying to fill a void."
Aug – Sep 2025
Project intern · remote
I extended an internal AI platform so automated compliance reports ran on a second CI system, and shipped it: audit-ready documentation generated in under 10 minutes instead of 40+ hours of manual writing. My PR merged to production.
40+ hours to under 10 minutes · merged to production
Jun – Aug 2025RCSI
Technical product manager · Dublin, on site
On site in Dublin I built the Clinical Skills Agent, a multimodal retrieval system for clinical education on Azure AI Search, Document Intelligence, FastAPI, and Vue. My LLM-driven build workflow cut spec-to-production time by roughly 90 percent, with mandatory review and human oversight as hard gates.
clinical AI shipped on site · SlideBot MVP in about five hours
2023 – 2025
Lead developer · intelligent remote observatories
I built the intelligent remote observatory stack: real-time satellite pass planning, automated slewing and capture, and a tethering capability that lets one operator coordinate thousands of networked telescopes. Before that I automated a New Mexico observatory's dome operations and imaging pipelines for Space 4 All.
thousands of telescopes, one operator
2023 – 2024RCSI
Lead developer, HealthNexAI · remote, Ireland
I led an AI education platform for clinicians from concept to deployment: a four-module tutor bot on GPT-4 with an Azure-hosted avatar experience for delivery and progress tracking. It bridged working clinicians to AI tooling they actually used.
led concept to deployment
2022 – 2024
Product manager, three stints · San Francisco Bay Area
I led Kohl's OpenAI onboarding as an intern, sitting across from their team and repeatedly beating them to their own roadmap. The exec chatbot I shipped answers leadership questions to this day, and I presented Marketplace Returns discovery to more than 1,000 employees and executives.
exec chatbot 2+ years in production
Sep – Nov 2023
Phortse, Nepal
Project lead · Khumbu, Nepal
I lived on site in a Sherpa village to co-design a community hospital and a waste-line system that improved water safety, coordinating Charity:Water, NEWAH, and local partners. Medical logistics stopped being abstract to me at 13,000 feet.
hospital and waterline co-designed on site
Summer 2018
Project intern · San Francisco
I paired on production client teams at fourteen, including StubHub, inside Pivotal Labs' XP culture. Outcome roadmaps, user research, and working software over claims got installed early, and they stuck.
first paired at fourteen
Agent operations
The multi-agent operation that runs my life in production: an orchestrator dispatching eight domain agents through custom MCP servers, a permission gate on every outward action, and control surfaces that are physical (a Stream Deck), remote (a two-way iMessage channel), and spoken (a wake-word satellite).
8 domain agents · 15+ tool subsystems
bed-remote daemon running since Aug 1
zero ungated outward actions, by standing rule
Agent-native hardware
A smart-home hub where the AI never claims an effect it did not witness on two independent channels. Ownership is a witnessed ritual nobody can forge over the air, census data is treated as untrusted input, and the LLM is never in the device hot path.
315 tests on real radios · M0 to M4 proven
21 devices re-recognized at 1.00 confidence, zero LLM tokens
live hub: 365 entities · 84 devices
AI verification
Independent, frozen, reproducible verification of AI claims. Hypotheses and pass thresholds are frozen to SHA-256 before any result exists, every artifact is content-addressed, and verdicts can be re-verified hash by hash by an auditor I have never met.
200-probe suite · daily CI evidence since Jul 16
RFC 3161 timestamp witness on every run
11 of 11 preregistered thresholds GO · independently replicated
Applied research
A semantic translation layer for enterprise SaaS that catches the fields APIs silently drop: build an independent reference corpus of what should exist, diff it against what the tenant actually returns. The absence is the signal.
404K rows · 47 tables · 150+ public sources
ghost-field detection patent filed Feb 2026, sole inventor
I am in New York, and I am looking for forward-deployed-shaped work: embedded with a customer, writing code on their real problems, judged on whether the thing runs in production. I am available part-time now and full-time from November 2026, when I finish my degree at NYU. If you are building agent infrastructure, verification tooling, or deployments that have to survive contact with the real world, I would like to talk.
crtmcvino1@gmail.com