Systems I've designed and built

A reference page, not a pitch: every system in one table, then one block per system with the problem it solved and what was done. Three are paid client systems in production. The rest are labelled for what they are. Every number was counted on the system itself and names its record; nothing is estimated, and there are no time-saved or revenue figures because none were measured.

On scope and honesty. Each build is a solo build, with Claude Code as the build tool. The agency client is anonymized at their request and its screenshots show sample data. The five GoHighLevel funnels are practice builds on my own account, not client work. The GoHighLevel workflow library is designed, not yet built inside GoHighLevel.

At a glance

Every system in one table

28 systems: 3 client, 2 production, 6 practice, 17 demo. Click a name for its block.

  • Clientpaid client work, in use by the client
  • Productionin real use on a schedule, my own
  • Practicemy own build to learn a platform, no client
  • Demobuilt to show the pattern, runs on request
SystemTypeToolsWhat it doesNumbers, sourced onlyLink
Client systems in production
Agency Command Center (Meta Ads + GoHighLevel)
Client
Dashboard app + 24 n8n workflows Next.js · TypeScript · Supabase · n8n Cloud · Meta Marketing API · GoHighLevel API · ClickUp · Netlify Reads every client's spend, billing, leads and speed to lead from Meta and GoHighLevel each morning and puts the agency's whole book on one screen. A failed read shows as a failure, never as a zero. about 50 clients (51 sub-accounts on the first scheduled sync) 24 active n8n workflows (n8n API list, 2026-11-02) 108 ad accounts inventoried (stack audit) 64 of 64 Meta ad accounts read clean (first live read) 72 of 73 GoHighLevel sub-accounts read clean (first full sweep) old feed dead about 6.5 months while runs reported success; old dashboard failing 31 of 38 CRM reads (audit, measured from the client's own sheet and dashboard)
Callixa: AI receptionist landing page
Client
Landing page, static build HTML/CSS/JS · Design tokens · Vercel · GoHighLevel slots (form, calendar, CRM) · gpt-4.1 translation dictionaries A conversion page where each visitor watches a call from their own industry get answered and booked. Seven client review rounds on live URLs, 45 languages, live on the client's own Vercel team. 7 review rounds (each deployed to its own URL) 11 industry verticals in the hero switcher (the page) 45 languages with a location-based default and right-to-left support (release of 2026-09-13)
WebPulse: scrape, score and rebuild engine + Studio
Client
AI pipeline (Claude Code kit) + review app Node.js · Playwright · Apify · Cheerio · GPT-4.1-mini vision · Claude Code skills · Vercel · Google Sheets Finds local businesses with outdated websites, scores them with a vision model, rebuilds the good leads as live preview pages with their own photos and colours, and delivers everything in the agency's Sheet. about $8 per 1,000 leads for the full pipeline (measured run, against a $30 budget) about 140 businesses scraped per city (Apify run per city)
Apps, sites and tools I own
FreshNest: cleaning-company site wired into GoHighLevel
Demo
Multi-page website + CRM wiring Astro · Vanilla CSS tokens · GoHighLevel forms + calendar · Vercel A hand-coded six-page cleaning site driven by one config file, with the quote form and booking calendar flowing into a GoHighLevel pipeline. 6 pages (Home, Services, About, Service Areas, Contact, Booking) 1 config file drives every page (the repo)
Trustworthy Cleaning: instant quote calculator
Demo
Quote calculator, static build HTML/CSS/JS · jsPDF · GoHighLevel inbound-webhook payload · Vercel A concept build for a real Austin cleaning company whose site promises a price in about a minute. Every answer moves a live price; the office view shows the lead score and the exact GoHighLevel payload. 6 answers move the price, each with its own bar (the page: service, bedrooms, bathrooms, condition, frequency, extras) 130 of 130 click-test checks on the live link (tests/click-test.mjs, 2026-11-09)
This site and its written door
Production
Static site + one Vercel function HTML/CSS/JS · Vercel static + serverless function · Node · nodemailer · Telegram Bot API The portfolio you are reading, with an animated system map and a "send the task" form whose function delivers each message to Telegram and Gmail and reports a status per channel. 2 delivery channels, each reporting ok, skipped or failed (the function's response) first real delivery 2026-11-07 (function response on the live site)
Voice-note-to-quote Telegram bot
Demo
Telegram bot, Python Python 3.12 · python-telegram-bot · OpenAI gpt-4.1 + gpt-4.1-mini · gpt-4o-mini-transcribe · Playwright PDF · SQLite A voice note or typed request to a private bot becomes a branded PDF quote. Reply to revise; /approve archives the version and drafts the client email. first draft about 10-15 s warm, 30-50 s cold (bot logs) revision about 5-8 s; approve + email draft about 5 s (bot logs)
Video to Notion knowledge base pipeline
Demo
Batch pipeline + Notion publisher Python · yt-dlp · ffmpeg · Gemini structured output · Embeddings + clustering · Notion API 2025-09-03 · SQLite manifest Watches a video library, keeps the frames the model says matter, merges the same idea taught across videos into one entry, and publishes four linked Notion databases. 10 videos, about 4.3 hours, in the demo run (pipeline manifest) 227 rows out across 4 linked databases: 65 concepts, 85 captioned screenshots (the published Notion workspace) $1.29 AI cost for the watch pass, 102 input tokens per second of footage (API usage, measured)
Meta static ads: template engine
Demo
Template pipeline, Node Node.js · Blade-subset compiler · Playwright renderer · Embedded Inter · gpt-image-2 cutouts Turns a reference ad into a validated template JSON that renders real text over product scenes at 1080 by 1350, gated by a build step that rejects anything that would fail review. 4 template modules; 11 rendered creatives at 1080 x 1350 plus a contact sheet (folder count, 2026-11-08)
Static Ads Factory
Demo
Image-generation CLI Node.js · Kie AI nano-banana-2 / pro · OpenAI gpt-image-2 · Disk cache Product photo in, ad scene imagery out: 40 tagged scene prompts, automatic model routing, a cost gate, a disk cache and retry. Typography is composited downstream by the template engine. 40 scene prompts in 6 buying presets (prompt library files, counted 2026-11-08)
GoHighLevel workflow library (11 specs)
Practice
Automation design: specs, registry, diagrams, emails. Designed, not yet built in GHL. GoHighLevel workflows, pipelines, custom fields, tags · Mermaid diagrams · HTML email One router and thin per-brand arms serving six funnels from a single sub-account: intake and auth, routing, quote engines, appointment lifecycle, welcome sequence, reviews, failure handling. Designed and documented; not yet built inside GoHighLevel. 11 workflow specs, each with a diagram (the specs folder) 21 canonical custom fields, 3 pipelines, 51 tags (the registry) 34 email templates built (the templates folder)
OLJ Watch: job-ad scanner
Production
Scheduled scanner to Telegram Node.js · Windows Task Scheduler · Telegram Bot API · No model calls Every 30 minutes, scans a job board on eight searches, scores each new ad against my profile and pay floor, and sends the match plus a ready application letter to my Telegram. runs every 30 minutes; 8 keyword searches; 72-hour posting window; $6 per hour floor (the tool's profile) first run: 201 ads scanned, 8 sent (run log, 2026-11-07)
Beautiful Websites Kit
Demo
Agent kit, 4 skills Node.js · Playwright · Claude Code skills · Vercel Point it at a niche and a city: it scrapes local businesses, scores each site for redesign potential, rebuilds the ones worth it with a different palette and layout each time, and deploys them. 4 skills; 2 sample rebuilds deployed (the kit and its live links)
n8n workflow library
Doc upload to Gemini File Store RAG Q&A + evaluation score
Demo
n8n workflow n8n Forms · Google Gemini File Store · OpenRouter (GPT-5-mini) · Google Sheets Upload a PDF once, auto-index it into a Gemini File Store, then chat for grounded answers with citations. An optional evaluation path scores answers and logs them to a Sheet. No counted figures
Google Maps prospect scraper + website contact enrichment
Demo
n8n workflow Apify · Firecrawl · Custom JavaScript · Google Sheets Pulls Google Maps listings into a Sheet, then reads each business website for emails and social links, writing a contact-ready row and skipping anything already processed. No counted figures
Instagram DM chatbot
Demo
n8n workflow Instagram Graph API · OpenAI (Whisper + agent) · Google Docs FAQ · Google Calendar Answers Instagram DMs, text or voice, from a FAQ document with conversation memory, checks the next 14 days of calendar availability, and replies on Instagram. availability checked over the next 14 days (workflow setting)
Telegram voice executive assistant
Demo
n8n workflow Telegram · OpenAI · Gmail · Google Calendar · Google Sheets Text or voice requests to a Telegram bot are transcribed, then an AI agent carries them out with connected tools: contact lookup in Sheets, sending through Gmail, booking in Google Calendar. No counted figures
AI virtual try-on generator for WooCommerce
Demo
n8n workflow WooCommerce · Google Sheets · Nano Banana image API · Google Drive Model and product images listed in a Sheet become a rendered try-on image; the job is polled to completion, saved to Drive, written back, and the WooCommerce product updated. No counted figures
Sales agent system
Demo
n8n workflow, 4 flows n8n · OpenRouter (Claude Sonnet 4.5) · Google Sheets CRM · Gmail · Calendly Four coordinated flows on one Sheets CRM: lead intake with IDs and dedupe, AI-personalized follow-ups capped at three, a booking listener that flips status instantly, and a no-show sweep every six hours. 4 flows; follow-ups capped at 3; no-show sweep every 6 hours (workflow settings)
Fathom-to-proposal generator (Google Slides)
Demo
n8n workflow Fathom webhook · Google Sheets · Google Slides · Google Drive · Gmail · OpenRouter (Claude Sonnet 4.5) A recorded sales call becomes a ready-to-review proposal deck: transcript cleaned, sections extracted by a model, a Slides template filled, the link emailed for review, the CRM row updated. No counted figures
HighLevel pipeline stage to Notion clients sync
Demo
n8n workflow HighLevel webhook · Notion The moment an opportunity changes stage in HighLevel, a webhook finds the matching client in Notion by contact ID and updates its pipeline stage in real time. No counted figures
AI contract automation (closed deal to PandaDoc to onboarding)
Demo
n8n workflow Google Sheets · Fathom · OpenRouter (Claude Sonnet 4.5) · PandaDoc · Calendly · Google Drive · Gmail A "Closed" update in Sheets becomes a PandaDoc contract with terms extracted from the call transcript; signatures trigger a welcome email, and an onboarding booking creates and shares the client's Drive folder. No counted figures
Notion business-ops suite (daily AI exec report + overdue escalation)
Demo
n8n workflow, 2 scheduled flows Notion · Gmail · ChatGPT Every weekday at 5 PM, Notion tasks become a KPI report with an AI summary emailed to the CEO; every four hours a monitor reminds assignees of past-due tasks and escalates after three ignored reminders. weekday report at 5 PM; sweep every 4 hours; one reminder per assignee per day; escalation after 3 (workflow settings)
GoHighLevel funnels
AdFlow Performance funnel
Practice
GoHighLevel funnel GoHighLevel funnels · Forms + tags A multi-step agency funnel: an audit landing page, a proposal step and an onboarding sequence. Practice build on my own subdomain. No counted figures
Pacific Coast Solar funnel
Practice
GoHighLevel funnel + custom JS GoHighLevel funnels · Custom JavaScript · Forms A solar landing page with an interactive savings calculator that estimates savings from roof size and energy bill before asking for contact details. Practice build. No counted figures
Pooch Scoop funnel
Practice
GoHighLevel funnel + custom JS GoHighLevel funnels · Custom JavaScript · Forms + calendar An opt-in funnel where a zip-code check filters leads by service area before booking is offered. Practice build. No counted figures
Dental Haven funnel
Practice
GoHighLevel funnel GoHighLevel funnels · Forms A dental opt-in funnel written for anxious patients: reassuring copy and visuals, and an intake form that asks only what the practice needs. Practice build. No counted figures
Legacy Shield Law funnel
Practice
GoHighLevel funnel + sequence GoHighLevel funnels · Workflows (email sequence) An estate-planning law funnel: an opt-in page plus a post-consultation sequence that explains the planning tiers. Practice build. No counted figures

Numbers come from the systems themselves: run logs, API inventories, audits, published records and folder counts. The record is named in brackets after each figure and again in the system's block below.

Group 1

Client systems in production

Three paid engagements. The agency client is anonymized at their request; every screenshot of its app shows sample data.

Agency Command Center (Meta Ads + GoHighLevel)

Client

Type Dashboard app + 24 n8n workflowsIndustry Meta-ads agency, home-service contractors

Next.jsTypeScriptSupabasen8n CloudMeta Marketing APIGoHighLevel APIClickUpNetlify

Description

A command center for a Meta-ads agency with about 50 home-service clients. A Next.js app on Supabase, fed by 24 n8n workflows on the agency's own n8n Cloud that read Meta and GoHighLevel each morning and in real time, write signed rows to one store, and alert the team where it already works. Everything runs on accounts the agency owns.

Problem it solved

The agency's reporting pipeline had been silently dead for about six and a half months while every run reported success, it covered roughly 36 of 108 ad accounts, and the dashboard the team used was failing 31 of its 38 CRM reads. The owner asked for a dashboard that is genuinely good, not a repair job, and ruled out a generic SaaS look.

What was done

Audited all 108 ad accounts and 73 sub-accounts first. Built the daily and real-time syncs (Meta, GoHighLevel, lead events, call-centre intake, lead quality, posting gaps), a signed ingest that checks signature, timestamp and schema version before any write, an error rail that files a failure row on any crash, the alert rail (client channel, DM, task board, delivery ledger) and ten app screens built around the daily check. Colour marks facts; a verdict appears only when the owner's own rulebook sets the target. On the GoHighLevel side: a lead-quality pipeline on every active account and an AI lead mover rolled out per account with the owner's OK. The first version ran on the agency's accounts three weeks in.

Numbers about 50 clients (51 sub-accounts on the first scheduled sync) · 24 active n8n workflows (n8n API list, 2026-11-02) · 108 ad accounts inventoried (stack audit) · 64 of 64 Meta ad accounts read clean (first live read) · 72 of 73 GoHighLevel sub-accounts read clean (first full sweep) · old feed dead about 6.5 months while runs reported success; old dashboard failing 31 of 38 CRM reads (audit, measured from the client's own sheet and dashboard)

Callixa: AI receptionist landing page

Client

Type Landing page, static buildIndustry AI receptionist for service businesses

HTML/CSS/JSDesign tokensVercelGoHighLevel slots (form, calendar, CRM)gpt-4.1 translation dictionaries

Description

A hand-coded landing page for an AI receptionist sold to trades, dental, legal and medical practices: a dark token-driven design system, a per-industry video hero with an 11-vertical switcher, a fully visible FAQ grid, plain-language pricing, and a language picker with 45 languages and a location-based default.

Problem it solved

The buyers are non-technical owners comparing the product with a human receptionist. The page had to demonstrate the product answering their kind of call rather than describe it, and the client's GoHighLevel page builder could not reproduce the fonts, tokens or components the design needed.

What was done

Turned the scope document into tokens and a ten-section page; ran four concept rounds and seven numbered client review rounds, each deployed to its own URL so the client compared live builds instead of screenshots; split the brand indigo into a decorative accent and an AA-safe text tint; proposed and shipped the page as a standalone static build with GoHighLevel supplying only form, calendar, CRM and tracking through marked slots; added the 45-language layer with generated dictionaries, a location-based default and right-to-left support; moved the deploy to the client's own Vercel team on their domain.

Numbers 7 review rounds (each deployed to its own URL) · 11 industry verticals in the hero switcher (the page) · 45 languages with a location-based default and right-to-left support (release of 2026-09-13)

WebPulse: scrape, score and rebuild engine + Studio

Client

Type AI pipeline (Claude Code kit) + review appIndustry Web design agency prospecting local businesses

Node.jsPlaywrightApifyCheerioGPT-4.1-mini visionClaude Code skillsVercelGoogle Sheets

Description

A Claude Code-native kit for a web design agency: composable skills chain scrape, qualify, extract assets, build, deploy and deliver, with one autopilot command per city. WebPulse Studio puts the pipeline behind a login so the team reviews each AI-built draft, fixes copy, photos and logo, and redeploys from one screen.

Problem it solved

The agency's best pitch is a finished redesign of the prospect's own site, but doing that by hand costs hours per prospect when cold outreach needs thousands. The client turned down an n8n build and wanted to drive everything through conversation.

What was done

Built the six-stage pipeline with cost discipline: AI only for the judgment calls (vision qualification, photo scoring), deterministic code for everything mechanical (brand extraction, template fill, palette repaint, deploy). Pre-flags skip franchises and no-website listings before any paid call. One Vercel project per niche with a route per lead. Delivery lands in the client's Google Sheet with preview URLs and optional outreach drafts; automatic sending was left out on purpose.

Numbers about $8 per 1,000 leads for the full pipeline (measured run, against a $30 budget) · about 140 businesses scraped per city (Apify run per city)

Group 2

Apps, sites and tools I own

Full builds that run for real or on request. Each one is labelled for what it is.

FreshNest: cleaning-company site wired into GoHighLevel

Demo

Type Multi-page website + CRM wiringIndustry Home services, residential cleaning

AstroVanilla CSS tokensGoHighLevel forms + calendarVercel

Description

A sample cleaning-company website built as a repeatable system: every headline, service, review, service area, brand colour and GoHighLevel embed lives in one config file, and the components read from it. Fictional business, sample content, no invented review numbers.

Problem it solved

Cleaning companies need a fast, mobile-first site that drops every visitor into the CRM without a page builder, and an agency needs to launch the next one without rebuilding from scratch.

What was done

Built the six pages in Astro with scoped design tokens and a few kilobytes of hand-written JS; wired the quote form and booking calendar to GoHighLevel; wrote the client setup guide and the launch steps (config, assets, embeds, build, deploy, DNS); removed invented ratings in an honesty pass and added indexable privacy and terms pages.

Numbers 6 pages (Home, Services, About, Service Areas, Contact, Booking) · 1 config file drives every page (the repo)

Trustworthy Cleaning: instant quote calculator

Demo

Type Quote calculator, static buildIndustry Home services, residential cleaning

HTML/CSS/JSjsPDFGoHighLevel inbound-webhook payloadVercel

Description

A concept instant-quote tool on the business's own brand: palette, font, crew photos, services, checklist and service area come from their website, and the layout and motion are cloned from a named Dribbble reference (two cards, rolling price, range tick, driver bars). Every dollar is a labelled concept rule, because the business publishes no prices.

Problem it solved

The company's homepage promises "your price in about a minute, no walkthrough, no callback", then collects a lead form with no number behind it. The tool had to give the number, explain it, and hand the office a lead it can act on.

What was done

Built the three-step form and the live estimate card. Out-of-area ZIP codes get a straight answer, and offices or homes with 5+ bedrooms go to a person instead of a fake flat rate. Contact details unlock the exact figure, never the range, and the PDF breakdown matches the page. The payload uses the GoHighLevel workflow library's field names, with UTM attribution and timestamped SMS consent; sample mode sends nothing until a real webhook is pasted. A footer line marks it as a concept build.

Numbers 6 answers move the price, each with its own bar (the page: service, bedrooms, bathrooms, condition, frequency, extras) · 130 of 130 click-test checks on the live link (tests/click-test.mjs, 2026-11-09)

This site and its written door

Production

Type Static site + one Vercel functionIndustry My own

HTML/CSS/JSVercel static + serverless functionNodenodemailerTelegram Bot API

Description

A hand-coded static site (no framework, self-hosted fonts) with one serverless function behind the "send the task" form. The function validates the message, drops honeypot submissions, posts to a dedicated Telegram bot and sends a Gmail copy with the visitor's address as reply-to, then returns ok, skipped or failed for each channel.

Problem it solved

A bare mailto link does nothing on a laptop with no mail app, and a form that only emails can fail silently. The site needed a written door that works for everyone and tells me when a channel breaks.

What was done

Built the form with client-side checks and a mailto fallback for local copies and failures; wrote the function with per-channel results, so one request is the whole health check; kept secrets on the Vercel project only; added the animated system map component (inline SVG, reduced-motion fallback, off-screen pause, keyboard focus); set trailing-slash routing so relative links hold on the live domain.

Numbers 2 delivery channels, each reporting ok, skipped or failed (the function's response) · first real delivery 2026-11-07 (function response on the live site)

Voice-note-to-quote Telegram bot

Demo

Type Telegram bot, PythonIndustry Agencies and founders who quote services

Python 3.12python-telegram-botOpenAI gpt-4.1 + gpt-4.1-minigpt-4o-mini-transcribePlaywright PDFSQLite

Description

A private Telegram bot for one founder. Voice notes are transcribed, a model drafts the quote as JSON under a brand skill (voice guide, schema, pricing rules, examples), Jinja2 and headless Chromium render the PDF, and a reply-to loop applies the smallest change and re-renders. /approve archives the version and drafts the send-to-client email.

Problem it solved

A typed quote means opening the template, writing the summary in brand voice, structuring deliverables, pricing, terms and exporting, when the founder already knows what to say and could say it in a minute.

What was done

Designed the three-layer structure (workflows, agent skill, tools); built the transcription, draft, render, revise and approve paths with SQLite state in WAL mode; fixed a racy quote-ID generator with an atomic sequence table; moved blocking render and model calls off the bot's event loop; swapped the model provider mid-build in four files because the skill content was provider-agnostic. No hours-saved figure is claimed until it is measured.

Numbers first draft about 10-15 s warm, 30-50 s cold (bot logs) · revision about 5-8 s; approve + email draft about 5 s (bot logs)

Video to Notion knowledge base pipeline

Demo

Type Batch pipeline + Notion publisherIndustry Course creators and founders with video libraries

Pythonyt-dlpffmpegGemini structured outputEmbeddings + clusteringNotion API 2025-09-03SQLite manifest

Description

A deterministic, resumable pipeline built for a 245-video brief and proven on ten public startup talks: acquire, proxy, transcribe (captions first), one structured watch pass per video, settled-frame capture around the moments the model flagged, embedding-based consolidation across videos, and an idempotent Notion publisher.

Problem it solved

Blind scene detection produces hundreds of unlabeled frames per video, 245 summaries are not a knowledge base, and a pipeline that cannot resume gets re-run from scratch a dozen times at that scale. The buyer also assumed model cost was the constraint; measured, it was not.

What was done

Wrote the stages as standalone scripts sharing one SQLite manifest; designed against Notion's real limits (data-sources model, about 3 requests per second, block and payload caps, three-call image uploads); built a pre-flight that checks token, sharing and write access step by step; measured the real token cost; published the library with schema, relations and uploads through the API. Video acquisition and caption parsing reuse an MIT tool credited in the repo; the orchestration, watch pass, capture, consolidation and publisher are original.

Numbers 10 videos, about 4.3 hours, in the demo run (pipeline manifest) · 227 rows out across 4 linked databases: 65 concepts, 85 captioned screenshots (the published Notion workspace) · $1.29 AI cost for the watch pass, 102 input tokens per second of footage (API usage, measured)

Meta static ads: template engine

Demo

Type Template pipeline, NodeIndustry DTC brands and performance-creative teams

Node.jsBlade-subset compilerPlaywright rendererEmbedded Intergpt-image-2 cutouts

Description

Built as the work sample for a high-volume static-ads role where templates are pasted into the client's Laravel system. A conversion prompt turns a reference image into template JSON; the build step validates schema, copy lengths and defaults and compiles the template; the renderer produces PNGs at the declared size with fonts and images inlined so output never drifts between machines.

Problem it solved

Payment on approved briefs means a rejected template is unpaid work plus a round trip, and most rejection causes are mechanical: undeclared variables, overlong default copy, malformed defaults, templates that do not compile.

What was done

Wrote the Blade-subset compiler, the validator, the Playwright renderer with a contact sheet, and an alpha probe that confirms product cutouts are truly transparent; authored the template archetypes and six briefs. Scope kept narrow on purpose: the subset compiles only what production Blade also compiles.

Numbers 4 template modules; 11 rendered creatives at 1080 x 1350 plus a contact sheet (folder count, 2026-11-08)

Static Ads Factory

Demo

Type Image-generation CLIIndustry DTC brands

Node.jsKie AI nano-banana-2 / proOpenAI gpt-image-2Disk cache

Description

The imagery half of a two-stage pipeline. A library of 40 scene prompts in six categories, selectable through six buying presets a media buyer speaks, is run per product; prompts that need in-image text route to the stronger text model; a dry run prices a batch before any spend; a cache and retry make re-runs cheap.

Problem it solved

Scene work and typography have different best models and different costs, and baked-in copy turns every headline test into a re-render.

What was done

Built the prompt library with tags and placement rules (Meta feed, Amazon main image white-only, A+ slots), the packshot helper that normalizes CDN images, routing by text need, the cost gate, the review sheet and offline smoke tests; documented the placement traps, such as off-white not being white.

Numbers 40 scene prompts in 6 buying presets (prompt library files, counted 2026-11-08)

GoHighLevel workflow library (11 specs)

Practice

Type Automation design: specs, registry, diagrams, emails. Designed, not yet built in GHL.Industry Home services, pet services, estate law (six practice brands)

GoHighLevel workflows, pipelines, custom fields, tagsMermaid diagramsHTML email

Description

A reusable automation layer for the six practice funnels: a closed vocabulary of real GoHighLevel steps, a canonical 21-field lead schema, three pipelines by commercial motion, colon-namespaced tags, custom values, six field maps, and eleven workflow specs, each with a Mermaid diagram checked by a structural validator.

Problem it solved

Six funnels collected six different field sets with inconsistent names, so no single router could attach to them, and an agency that re-clones a snapshot per client never ends up with a library.

What was done

Wrote the registry and the eleven specs (intake and auth, lead router, two quote engines, order and session, appointment lifecycle, welcome sequence, speed-to-lead and missed-call text-back marked off, review request with a rating gate, ops and failure handler); built 34 branded email templates on one shell; decided against A2P registration for a fictional brand and left every SMS step built and disabled. The build inside GoHighLevel is the next phase and is not claimed.

Numbers 11 workflow specs, each with a diagram (the specs folder) · 21 canonical custom fields, 3 pipelines, 51 tags (the registry) · 34 email templates built (the templates folder)

OLJ Watch: job-ad scanner

Production

Type Scheduled scanner to TelegramIndustry My own

Node.jsWindows Task SchedulerTelegram Bot APINo model calls

Description

A personal tool: eight keyword searches, a scoring rubric (platform skills, dashboards, automation, agency employer, role shape), hard blockers for roles I do not want, a pay parser that converts monthly and peso figures to an hourly rate, and a letter assembled from measured proof blocks with the portfolio link. A seen-list makes sure nothing is sent twice.

Problem it solved

Applying within hours of a posting is the edge on that board, and reading every new ad by hand all day is the opposite of automation.

What was done

Built the scanner, the scoring and filters (tuned against nine of my own accepted and rejected examples), the Telegram sender and the hidden scheduled task; added a check mode that grades a single ad with reasons, a dry run, and a stateless cloud mode for a hosted routine.

Numbers runs every 30 minutes; 8 keyword searches; 72-hour posting window; $6 per hour floor (the tool's profile) · first run: 201 ads scanned, 8 sent (run log, 2026-11-07)

Beautiful Websites Kit

Demo

Type Agent kit, 4 skillsIndustry Local services (nail salons and med spas in the sample runs)

Node.jsPlaywrightClaude Code skillsVercel

Description

My own tooling, kept generic: four skills that run inside an agent IDE, with cost rules and file conventions in the kit's own instructions, and a build log that records each site's palette and layout so the next one differs.

Problem it solved

Pitching a redesign works best with the prospect's own site rebuilt, and a kit that produces the same template twice is a liability.

What was done

Wrote the four skills and the workflow that chains them, the cost rules, and the build log; deployed two sample rebuilds as proof.

Group 3

n8n workflow library

Ten workflows with their canvases on the automation page. Built and runnable on my own n8n; demo on request. No client deployment is claimed for these.

Doc upload to Gemini File Store RAG Q&A + evaluation score

Demo

Type n8n workflowIndustry Any team with policy or product documents

n8n FormsGoogle Gemini File StoreOpenRouter (GPT-5-mini)Google Sheets

Description

An n8n form accepts a PDF and indexes it into a Gemini File Store; a chat flow answers questions grounded in that store with citations. An optional evaluation path scores answers and logs them to Google Sheets.

Problem it solved

A team's policies and product documents live in PDFs nobody searches, and a chatbot that answers from general knowledge invents things.

What was done

Built the upload and indexing path, the grounded Q&A with citations, and the evaluation logging so answer quality can be tracked over time instead of trusted.

Numbers No counted figures. Nothing here is estimated.

Google Maps prospect scraper + website contact enrichment

Demo

Type n8n workflowIndustry Lead generation for local-service outreach

ApifyFirecrawlCustom JavaScriptGoogle Sheets

Description

Apify pulls Google Maps listings for a niche and city into a Sheet; Firecrawl then reads each business website for emails and social links (LinkedIn, Facebook, Instagram, X), writes a contact-ready row, and skips anything already processed.

Problem it solved

Outreach lists bought or scraped by hand are stale, duplicated, and missing the contact detail that makes a first message possible.

What was done

Built the scrape-to-Sheet flow, the per-site enrichment with custom JavaScript parsing, and repeat-prevention so re-runs only add new businesses.

Numbers No counted figures. Nothing here is estimated.

Instagram DM chatbot

Demo

Type n8n workflowIndustry Appointment-based local businesses

Instagram Graph APIOpenAI (Whisper + agent)Google Docs FAQGoogle Calendar

Description

Instagram DMs arrive by webhook; echo messages are filtered, voice notes are transcribed with Whisper, and an AI agent grounded on a Google Docs FAQ with conversation memory answers questions and checks the next 14 days of Google Calendar availability, replying on Instagram.

Problem it solved

Appointment-based businesses lose DMs that arrive after hours or as voice notes, and a generic bot cannot answer in the business's own policies.

What was done

Built the webhook intake with echo filtering, the text-versus-voice routing and transcription, the FAQ-grounded agent with memory, the calendar availability tool and the reply path.

Numbers availability checked over the next 14 days (workflow setting)

Telegram voice executive assistant

Demo

Type n8n workflowIndustry Founders and executives

TelegramOpenAIGmailGoogle CalendarGoogle Sheets

Description

Text or voice messages to a Telegram bot are transcribed, then an AI agent carries out the request with connected tools: contact emails from Google Sheets, sending through Gmail, booking in Google Calendar, with conversation memory and an immediate acknowledgement.

Problem it solved

Founders dictate tasks on the move; nothing turns "email Sam and book Thursday" into the email and the booking without a person in the loop.

What was done

Built the Telegram intake, the transcription, the agent with three tools and memory, and the acknowledgement reply.

Numbers No counted figures. Nothing here is estimated.

AI virtual try-on generator for WooCommerce

Demo

Type n8n workflowIndustry E-commerce, apparel

WooCommerceGoogle SheetsNano Banana image APIGoogle Drive

Description

Model and product images listed in a Google Sheet are sent to an image model that renders the model wearing the product; the job is polled until complete, the image saved to Google Drive, the link written back to the Sheet, and the matching WooCommerce product updated.

Problem it solved

Try-on imagery for an apparel catalogue means a photoshoot per product; without it, listings show flat product shots.

What was done

Built the Sheet-driven batch, the Drive link conversion, the asynchronous job polling, the upload and write-back, and the WooCommerce product update.

Numbers No counted figures. Nothing here is estimated.

Sales agent system

Demo

Type n8n workflow, 4 flowsIndustry Sales teams

n8nOpenRouter (Claude Sonnet 4.5)Google Sheets CRMGmailCalendly

Description

Four coordinated flows sharing one Google Sheets CRM: lead intake with IDs and dedupe, AI-personalized follow-ups capped at three, a Calendly booking listener that flips status instantly, and a no-show sweep every six hours that sends a reschedule email and updates the row.

Problem it solved

Most lost deals were never rejected. The follow-up lived in someone's head, and booked-but-ghosted leads were written off.

What was done

Built the four flows and the shared CRM state so any flow can pause or fail without breaking the others; the model writes each follow-up from the lead's own context and stops after three.

Numbers 4 flows; follow-ups capped at 3; no-show sweep every 6 hours (workflow settings)

Fathom-to-proposal generator (Google Slides)

Demo

Type n8n workflowIndustry Agencies and consultants

Fathom webhookGoogle SheetsGoogle SlidesGoogle DriveGmailOpenRouter (Claude Sonnet 4.5)

Description

A recorded sales call arrives by webhook from Fathom; the workflow pulls the lead's details from Sheets, cleans the transcript, extracts proposal sections (title, bottlenecks, phases, pricing) with a model, copies a Google Slides template, fills the placeholders, emails the deck link for review and updates the CRM row.

Problem it solved

Writing the proposal deck after a discovery call is the same hour of work every time, and the details are already in the recording.

What was done

Built the webhook capture, the transcript cleanup, the AI section extraction, the Slides template copy-and-replace, the review email and the CRM update.

Numbers No counted figures. Nothing here is estimated.

HighLevel pipeline stage to Notion clients sync

Demo

Type n8n workflowIndustry Agencies running delivery in Notion

HighLevel webhookNotion

Description

The moment an opportunity changes stage in HighLevel, an n8n webhook finds the matching client in Notion by GoHighLevel contact ID and updates its pipeline stage.

Problem it solved

The CRM and the delivery team's workspace disagree about where a client stands, because somebody has to re-type the stage.

What was done

Built the stage-change webhook, the lookup by contact ID and the Notion update.

Numbers No counted figures. Nothing here is estimated.

AI contract automation (closed deal to PandaDoc to onboarding)

Demo

Type n8n workflowIndustry Agencies

Google SheetsFathomOpenRouter (Claude Sonnet 4.5)PandaDocCalendlyGoogle DriveGmail

Description

A "Closed" update in Google Sheets becomes a PandaDoc contract: key terms are extracted from the Fathom transcript with a model, a review link is sent, the contract ID and status are logged to the CRM; a signature triggers a welcome email, and a Calendly onboarding booking creates and shares the client's Google Drive folder.

Problem it solved

Between "yes" and a signed contract, deals stall on paperwork assembled by hand from a call everyone has already had.

What was done

Built the trigger, the AI term extraction, the PandaDoc document creation, the signature monitoring, the welcome email and the onboarding folder automation.

Numbers No counted figures. Nothing here is estimated.

Notion business-ops suite (daily AI exec report + overdue escalation)

Demo

Type n8n workflow, 2 scheduled flowsIndustry Ops teams

NotionGmailChatGPT

Description

Two scheduled flows: every weekday at 5 PM, Notion tasks (done today, overdue, blocked, waiting on client) become a KPI report with an AI executive summary saved to a reports page and emailed to the CEO; every four hours a monitor sweeps past-due tasks, reminds each assignee at most once a day and escalates to the owner after three ignored reminders.

Problem it solved

Ops teams drift when nobody reads the board, and the owner finds out about overdue work in a meeting.

What was done

Built the daily report with the model summary, the 4-hour sweep with per-assignee reminder limits, and the escalation rule.

Numbers weekday report at 5 PM; sweep every 4 hours; one reminder per assignee per day; escalation after 3 (workflow settings)

Group 4

GoHighLevel funnels

Five practice builds on my own GoHighLevel sub-account and subdomains. Not client engagements, and no traffic figures are claimed.

AdFlow Performance funnel

Practice

Type GoHighLevel funnelIndustry Marketing agency

GoHighLevel funnelsForms + tags

Description

A multi-step agency funnel built in GoHighLevel on my own subdomain: an audit landing page, a proposal step and an onboarding sequence, with forms and tags carrying the lead between stages.

Problem it solved

Agencies lose prospects between a free audit and a signed proposal because each step lives in a different tool.

What was done

Built the three stages as one funnel with forms and tags carrying the lead between them. Practice build: no client, no traffic figures claimed.

Pacific Coast Solar funnel

Practice

Type GoHighLevel funnel + custom JSIndustry Solar installation

GoHighLevel funnelsCustom JavaScriptForms

Description

A solar landing page in GoHighLevel with an interactive savings calculator embedded as custom JavaScript, estimating savings from roof size and energy bill before the contact form.

Problem it solved

Solar leads bounce when the first screen asks for a phone number; a calculator gives value first.

What was done

Wrote the calculator and embedded it in the GoHighLevel page with the lead captured after the estimate. Practice build on my own subdomain.

Pooch Scoop funnel

Practice

Type GoHighLevel funnel + custom JSIndustry Pet waste removal

GoHighLevel funnelsCustom JavaScriptForms + calendar

Description

An opt-in funnel for a pet-waste-removal service where a zip-code check filters leads by service area before the booking step is offered.

Problem it solved

A booking from outside the service area wastes a route and ends in a refund.

What was done

Built the qualification logic in the page and the GoHighLevel form and booking flow behind it. Practice build.

Dental Haven funnel

Practice

Type GoHighLevel funnelIndustry Dental practice

GoHighLevel funnelsForms

Description

A dental opt-in funnel in GoHighLevel written for anxious patients: reassuring copy and visuals, and an intake form that asks only what the practice needs.

Problem it solved

Dental pages that lead with procedures scare off the patients who most need to book.

What was done

Wrote the copy and built the page and intake form in GoHighLevel. Practice build.

Legacy Shield Law funnel

Practice

Type GoHighLevel funnel + sequenceIndustry Estate-planning law

GoHighLevel funnelsWorkflows (email sequence)

Description

An estate-planning law funnel in GoHighLevel: an opt-in page and an automated post-consultation sequence that walks the client through the planning tiers.

Problem it solved

Clients who book one consultation rarely learn what else the firm offers, so the firm never gets to offer it.

What was done

Built the opt-in page and the post-consultation education sequence. Practice build; no effect on case value is claimed.

Available for remote work

Want the system behind one of these?

Bring the task your team repeats every week to a free 30-minute call, or send it in writing. I'll say honestly whether it is worth automating, what I'd build, and roughly what it costs.