Pat McGovern

Name:

PAT McGOVERN

Base:

NEW HAVEN, CT

About:

I build software for real places and real problems. Civic data maps, live bird tracking, AI agents that know when to stop. Everything below is live, public, and in use. I built each one myself.

I spent years investing at Bowery Capital, and before that led Palantir deployments for law firms and public sector clients. Now I build. The pattern across everything here: find a system people are stuck with, a phone tree, a records portal, a spreadsheet, and replace it with software that respects their time.

Contents:
  1. 01NHV Navigator
  2. 02Flockline
  3. 03Elm City Explorer
  4. 04Portal Operator
  5. 05Vertical Velocity
Links:

GitHubEmail

01

NHV Navigator

Social services, minus the phone tree

What:

A free, multilingual directory of social services in New Haven: food, housing, healthcare, legal aid, immigration. Replaces the 211 phone tree with search, an eligibility wizard, and an AI assistant that answers from the live database. Sister sites serve Bridgeport and Hartford.

Notes:
  • 200+ service listings, counted live from the database
  • 7-step eligibility wizard with confidence scoring
  • Voice hotline and text-me-this-resource via Twilio
  • Three cities: New Haven, Bridgeport, Hartford
Stack:

Next.js / TypeScript / Prisma / Supabase / Claude / Twilio / Vapi

Proof:

6 languages served

Live:

nhvnavigator.com

NHV Navigator illustration of New Haven at dusk
Fig. 1 · NHV Navigator as deployed
02

Flockline

Live bird movement, on a map

What:

A live map of recent bird sightings across the United States, built on eBird data from the Cornell Lab. Scrub a timeline to watch movement day by day, keep a watchlist with alerts, and get a weekly regional email digest written from the actual data. Includes original field-guide artwork for the species pages.

Notes:
  • Nationwide coverage with four regional presets and state-level refinement
  • Daily timeline scrubber over live eBird observations
  • Weekly Roundup email digest, generated and sent automatically
  • Field and Dusk themes, original bird illustrations
Stack:

React / Vite / TypeScript / Leaflet / Express / Resend / Claude

Proof:

1,400+ species tracked

Live:

flockline.appsource

Flockline map and field-guide artwork
Fig. 2 · Flockline as deployed
03

Elm City Explorer

Who actually owns New Haven

What:

An interactive map of corporate-owned residential property in New Haven. Python scripts scrape the city assessor and the CT Secretary of State, then group the LLCs and shell companies into landlord networks, so a resident or a reporter can trace any building back to the people behind it.

Notes:
  • 400+ landlord networks identified by principal and agent matching
  • Weekly refresh with a what-changed filter
  • Per-network profile pages with parcel polygons
  • Built from public records: assessor rolls and state business filings
Stack:

Python / Leaflet / Vercel / Claude / CT SOS data

Proof:

7,000+ properties mapped

Live:

elmcityexplorer.com

Elm City Explorer property ownership map
Fig. 3 · Elm City Explorer as deployed
04

Portal Operator

An agent that knows when to stop

What:

An AI agent that operates a legacy insurance carrier portal in a real browser, with no API and no inside help. Each turn it sees a screenshot and the on-screen controls, then acts. The point is restraint: two of the four test scenarios are traps the agent has to refuse, without inventing a confirmation number.

Notes:
  • Two scenarios require the agent to refuse and write nothing
  • Every step recorded: screenshot, reasoning, action, outcome
  • Scored against the portal's own records, not the agent's claims
  • The full test run cost about $0.66
Stack:

Claude Sonnet 4.5 / Playwright / TypeScript

Proof:

4/4 scenarios passed

Live:

github.com/pwmcgovern-afk/portal-operator

Portal Operator agent mid-run in the carrier portal
Fig. 4 · Portal Operator as deployed
05

Vertical Velocity

The capital-efficiency scoreboard

What:

A leaderboard ranking vertical AI companies by ARR per employee, the number that actually separates efficient builders from headcount-heavy ones. Every company page carries headcount, revenue, valuation, founders, and milestones, and each company gets its own share card.

Notes:
  • Ranked by ARR per employee, not raw ARR
  • Per-vertical filters and an efficiency calculator
  • A share card for every company
  • Feeds a daily agent that watches all 84 companies for new job postings
Stack:

React / Vite / TypeScript / Framer Motion / @vercel/og

Proof:

84 companies ranked

Live:

verticalvelocity.co

Vertical Velocity leaderboard
Fig. 5 · Vertical Velocity as deployed
Writing:

Capital Efficient, my newsletter on building software companies without burning money. Essays on venture math, efficient teams, and what the numbers actually say.

Consulting:

Accipiter Labs, my consulting practice. I build AI systems inside real businesses, starting with a diagnostic of where the manual work actually is.