BUILDLOG
21 APR – 10 JUL 2026 · ~12 WEEKS
Georg S. · Solo operator · Germany

From zero coding to a production-grade automation system in twelve weeks.

A single person, working only with AI, built what would normally take a small team several months — with quality control, automation, and a sellable side-product. No prior software-engineering background.

430+
Git commits
300+
Local businesses analysed
142
Full SEO audits
580+
Personalised outreach emails
The headline

~500 hours of skilled work, compressed into a system that runs unattended.

One complete lead — finding it, auditing it, cross-checking five data sources, drafting a personalised approach, passing an 18-point quality gate — is roughly 3–4 hours of skilled human work by hand. The pipeline does it in minutes. Across 300+ analysed businesses and 142 full audits, that's the equivalent of months of full-time manual labour — built and run by one person.

The builds

Six things shipped, in order of weight

01

Local-SEO Outreach Pipeline the core engine

An end-to-end analysis and automation system for local beauty, wellness and TCM businesses. It runs automated lead research, builds a defensible SEO audit, and produces a personalised outreach draft — assembled in minutes, with the final review and release always handled by a human.

1 Lead research 2 5-source data gathering 3 Second-model cross-check 4 Draft creation 5 18-point quality gate 6 Roast pass 7 Email verification 8 Draft filing
Proof it's real, not a toy

430+ commits, 255 documented pipeline runs, 27 documented auto-aborts — the quality gate actively rejects weak drafts rather than letting them through. First reference client live: a Munich wellness studio that ChatGPT and Gemini now recommend at #1 for its money keyword (measured June 2026) — with a 5.0★ Google review for the work.

02

gego AI-visibility tracker · Go

A self-built command-line tool in Go that tracks how a brand appears in AI answers (ChatGPT, Perplexity, Gemini) — the "are we cited by the AI?" question classic SEO tools can't answer. Compiles to real binaries, runs weekly, ships with proper repo structure.

The rarest asset here

LLM-visibility tracking barely exists as a product in the German-speaking market. Built once, it now runs on its own — the only strand of the business that scales without costing more of my time.

03

blitztext voice-to-text · Python · free on request

One of the very first tools built (20 Apr 2026), hardened since. A global-hotkey speech-to-text tool — hold Ctrl+Shift, speak, text appears in any app. Two variants: a fully local build (faster-whisper on GPU, nothing leaves your machine) and a lightweight cloud build (bring your own OpenAI key). Windows. macOS users: check out Christoph Magnussen's open-source blitztext-app ↗ (Swift menubar app).

I'm giving it away — free

Dictation runs ~3× faster than typing; it paid for its own build time within days. If you want either variant (local or cloud), message me on Skool — link in the footer — and I'll send you the package with setup instructions.

04

Business Website live · full relaunch 09 Jul

A public multi-page business site — blog, industry pages, case studies, pricing, legal pages, an SEO glossary (31 terms) and a live AI-visibility case study. Fully relaunched on 9 July: 17 pages moved to a unified editorial design system with self-hosted fonts, iterated in an AI design tool and ported by the same AI workflow that built everything else.

An agency-scale deliverable, solo

A custom multi-section business site with a content pipeline behind it is typically a multi-week agency project costing several thousand euros. Built, redesigned and maintained alone — the full 17-page relaunch shipped in two days.

05

GEO Product + geo-audit Skills citability scoring

A complete GEO (Generative Engine Optimization) guide, versions 1 through 5, plus a suite of custom AI skills — geo-audit, geo-compare, geo-monitor, geo-fix-content, geo-fix-schema — that score a website's citability by AI engines across four dimensions.

Fuzzy judgment → repeatable score

What would be half a day of manual analysis per site becomes a structured skill run — and the same skills power both the consulting service and the tracker product.

06

BookingDealHunter deal scanner · scheduled

A fully automated Booking.com deal scanner: stealth scraping, deal parsing, database persistence, SHA-dedupe, Telegram push alerts, budget caps and a local config UI. Runs unattended four times a day on a scheduled task.

Hardened like production software

Rebuilt and hardened in a single day in July: connection lifecycle, single-instance locks, retry logic, price-median filtering — 105 automated tests green.

Also in the workshop
Semantic project memory
A local semantic-search memory over 1,000+ project documents, ingested daily — the AI assistant recalls decisions from any past session in seconds. Daily ingest optimised from 5.5 h to 2 min.
Heidi
TikTok / social AI-influencer project with a generative content pipeline. Research & build phase.
TikTokAutocutter
Automatic video-cutting prototype.
PainTracker
Tracking-tool prototype.
The bottom line

Twelve weeks, counted

430+
Git commits (pipeline repo)
300+
Local businesses analysed
142
Full SEO audits
580+
Personalised outreach emails
255
Pipeline runs · 27 auto-aborts
19+
Tools shipped
pipeline · CLI · trackers · gates
1
Reference client live
#1 in AI answers · 5.0★ review
105
Tests green on the newest build alone

In one sentence: a solo operator, starting from no engineering background, used AI to build a quality-controlled SEO analysis & automation infrastructure, a relaunched 17-page business site, and a sellable AI-visibility product in twelve weeks — work that would normally require a small team and a budget many times the cost.