will@gtm: ~

$ whoami

Will Mauro

GTM ENGINEER

I build the systems that find, research, score and reach the right customers, and the software that runs them: apps, dashboards, integrations and AI agents. Then I test it all until the numbers hold up.

thewillmauro@gmail.com GitHub

Upper East Side, New York · in-office or remote

Are you who I applied to?

This is a working terminal. Type your company's website and press Enter. I already did the research.

Welcome. Things you can do here:
> Type your company's website (like yourcompany.com) to open the project I built for you.
> Type help to see every command.

guest@will · case files: 1 · updated Nov 22, 2026

summary.txt

The 30-second version

85%right decision-maker found by my AI research agent
4.5¢research cost per company
20→26correct out of 30 for my AI agent team
  • Outbound research pipeline: finds companies, the right contact and fit, then syncs to the CRM and sends safely.
  • Resi3D: a 3D property-scanning business I built end to end. Bookings, Stripe invoicing, client portal, condition reports. Live in production.
  • AI agent team: nine assistants with hard limits. Nothing goes live, reaches a client or spends money without my approval.
  • Before that: 30+ enterprise accounts at Amplience, growth marketing, and first hire at a YC-backed startup.
cat what-broke.log

What broke, and what I changed

Most portfolios only show the wins. These are six real problems from my own systems.

Fixed

The 10/10 that wasn't real

ProblemMy research agent's first test run found the right contact at 10 of 10 firms.

CaughtA perfect single run looked too good, so I ran it again. Results varied.

FixI report the average of repeated runs (85%), grew the test set to 17 hand-labeled firms, and block any change that scores worse.

Fixed

Safety rules that were only suggestions

ProblemMy nine AI assistants had rules written into their instructions.

CaughtI tested whether the rules held. Most could be ignored.

FixHard limits: nothing goes live, reaches a client, or costs money unless I approve it.

Fixed

Practice answers leaking into memory

ProblemAnswers from practice tasks were mixing into the agents' record of real work.

CaughtThe test runs surfaced it.

FixTest runs now stay completely separate from real work.

Caught before release

A test that could email a real person

ProblemOn Resi3D, test copies of the site shared the live email account.

CaughtBefore shipping I asked: can a test booking reach a real person? It could.

FixTest bookings are flagged at creation and can only email our own inbox.

Fix in review

Refunds the system ignored

ProblemStripe sent refund events, but the code never read them, so a refund still showed as paid.

CaughtComparing the database against Stripe instead of trusting the dashboard.

FixA refund now updates the invoice, relocks downloads, and posts an alert.

Fix in review

A real lead who waited 12 days

ProblemA real booking request sat unanswered because nothing flagged it.

CaughtPulling real usage numbers instead of assuming things worked.

FixA speed-to-lead alert every 30 minutes, and time to first reply on the dashboard.

evals.log

How I measured it


      
ls ./built

What I've built

outbound-research-pipeline

Sep–Oct 2026 · built for Veroxa, a legal-tech app I ran
  • Found family law firms, the right person to contact, and a fit score, then delivered a ranked list into the CRM.
  • Filled in emails by checking 11 data providers in turn through Clay.
  • Synced to HubSpot without duplicates or overwriting manual edits; opt-outs honored everywhere.
  • Sent through Smartlead from warmed-up domains, with daily caps and an automatic pause if bounces spiked.
  • Nearly 500 automated tests, plus a check that blocks any change that makes accuracy worse.
Claude APIClayHubSpotSmartleadSupabaseTypeScript

resi3d

Nov 2026–now · a 3D property-scanning business · live in production
  • Booking site, quotes from the price list, Stripe invoices, prepaid packs, and a client portal.
  • Timestamped condition reports with photos, expiring share links, and a move-in vs move-out comparison.
  • Booking and payment alerts, a daily digest, and a daily list of anything slipping through the cracks.
SupabaseVercelStripeResendDiscord

agent-team --hard-limits

2026
  • Nine AI assistants, each with one job (such as engineering, sales or finance) and access to only the tools that job needs.
  • Safety rules turned into hard limits after testing showed most could be ignored.
Claude CodeMCPEvalsPermissions
cat rules.txt

How I work

Measure it honestly

Hand-labeled test sets, averaged runs, and a check that stops changes that make things worse.

People approve what matters

Agents draft and research. Anything that goes live, reaches a customer, or spends money waits for a yes.

Leave it explainable

Plain-language docs and a written record of every design decision.

history

Background

  1. 2019–2026Vantage Mode
    Founder. GTM engineering from Apr to Sep 2026; before that, photo, video and creative for brands, Shopify merchants and B2B SaaS companies.
  2. 2017–2019Amplience
    Customer Success Manager. 30+ enterprise accounts at $200K–$500K a year each, including Crate & Barrel, TUMI and Tom Ford.
  3. 2016–2017ALOHA
    Brand Marketing & Growth. Tracked AOV, LTV, CAC and retention for a Shopify subscription brand, and rebuilt lifecycle email with behavioral segmentation.
  4. 2015–2016House of Nutrition
    Ecommerce Marketing Manager. Led the move to BigCommerce and Amazon FBA, and ran sales across Amazon, eBay, Jet and web for 20,000+ SKUs.
  5. 2014–2015Two Tap (YC)
    Business Development & Product Marketing, first hire. Managed 200+ retailer and brand relationships for a universal shopping-cart API.
  6. 2013–2014Shopsy
    Co-Founder, Product Marketing. A platform for buying products seen in Instagram photos; closed partnerships with 100+ brands, including Sephora and Lululemon.
  7. 2012Manhattan College
    B.S., Computer Information Systems and Management.