Founder-fronted · axelbaum.com · Bay Area handyman niche
AXL Sites
A web-design business built on one rule: never ask a stranger to imagine anything.
Before any cold email goes out, a pipeline renders that business a complete sample site, on its own private URL, with their real name, city, and phone on it. The email is one sentence and a link to proof.
Best viewed in Internet Explorer
Handyman in San Francisco
Repairs done right, on time, at a fair price.
☎ Tap to callDrag. Illustrative wireframes, not a client’s real site; the engine composes with the prospect’s actual content.
The operation, in real numbers
Python orchestration over an Astro template: scout, diagnose, build, check, publish, pitch. June 2026 snapshot from the pipeline ledger.
00 scout
public listings → prospect folders
00 diagnose
50-word pitch angle per prospect
00 build
Astro render, 4 routes each
00 check
rubric gate: no agency-speak
00 publish
private noindexed URL
00 pitch
human-gated first touch
00 reply
first test batch
zero replies on the nine-send test batch is the honest baseline the instrumentation was built to measure.
Trades-to-trades, not agency-to-trades
The mockup has to read like an established local business’s site, because render quality maps directly to reply rate. Each render produces four real routes (home, about, services, contact) with the prospect’s name, service area, and phone number, published noindexed on my domain so it is private to the pitch. A rendered batch takes about thirty seconds per prospect.
Deliverability is engineered like the rest of it: a warmed founder domain, thirty sends a day maximum with minutes between them, bounce and complaint thresholds that pause everything, and a three-touch sequence where only the first email is written and gated by hand. Replies are classified automatically and a positive one stops the sequence cold.
Where it honestly stands
Pre-revenue. The first nine-prospect test batch ran the full sequence and produced zero replies; one custom engagement reached a proposal and declined before build. That is what the instrumentation is for: the pipeline is proven end to end, the pitch is now the variable under iteration, and the next batch goes out with the domain fully warmed. Every stage of that learning loop is logged.
Astro + Tailwind template · Python orchestration · Gmail-native threading · reply classifier (7 tiers) · DMARC digests · suppression lists · Vercel hosting on axelbaum.com