◆ Case Study · Marketing OS

One Brief In, Full-Funnel Campaign Out — Across 6 Channels, 3 Languages, Zero Ad Spend

GAEL is a self-hosted marketing operating system. Feed it a campaign brief and it generates, adapts, schedules, publishes, captures leads, and nurtures them — with n8n orchestrating DeepSeek, Listmonk, Mautic, Chatwoot and Umami on a single VM. No Zapier, no Make, no Mailchimp, no HubSpot.

Channels Orchestrated

20+

n8n Workflows

Languages

Unlimited

Email Capacity

Ad Spend

The 4 Campaign Templates

Template	Duration	Cadence	Use Case
Product Launch	14 days	Teaser → Reveal → Social proof → Urgency	New product, feature release, course drop
Evergreen	Continuous	3 pins/day, 1 post/day, 2 tweets/day, weekly newsletter	Always-on traffic to landing pages
Flash Sale	24–72 h	All channels, posts every 2–4 h, email blast + reminder	Promo, last-call, inventory clearance
Lead Magnet	Continuous	Free value → landing → auto-delivery → nurture	Ebook, checklist, webinar opt-in

New campaign types can be added as n8n sub-workflows in hours. The brief schema, content engine and distribution hub stay the same — only the cadence template changes.

Tech Stack

Orchestration

n8n20+ workflows

PostgreSQLcontent queue + CRM

Redisrate limits + cache

Telegram botops + alerts

Content + AI

DeepSeekcopy + chat AI

GPT Imagevisuals

Leon (Remotion)video render API

Google TTSvoiceover

Capture + Nurture

Mauticlanding pages + forms

Listmonkemail (36 lists)

Chatwootomnichannel inbox

Umamianalytics + funnels

System Architecture

GAEL is structured as a 6-layer pipeline. Every layer is self-hosted on a single Oracle Cloud ARM VM, orchestrated by n8n workflows, and state lives in PostgreSQL. No external SaaS dependencies for the core pipeline — only the channel APIs themselves.

Layer 1 · Input

Campaign brief webhook · scheduled triggers · Telegram commands

Layer 2 · Content Engine (n8n)

Campaign Router → DeepSeek text gen · GPT Image visuals · Leon video render · Mautic page gen

Layer 3 · Distribution Hub

Postgres content_queue → hourly publisher → Pinterest · Facebook · Twitter/X · Instagram · TikTok · Listmonk

Layer 4 · Capture (Mautic)

Landing pages per campaign · forms with UTM tracking · lead scoring by engagement

Layer 5 · Nurture

Listmonk drip campaigns (36 segmented lists) · Chatwoot omnichannel inbox with AI agent

Layer 6 · Intelligence

Umami funnels · engagement scoring · A/B test feedback loop → back into Layer 1

Core n8n Workflows

Workflow	Trigger	What it does
`campaign-launcher`	Webhook / manual	Receives brief, decides channels, kicks off content generation, starts email sequence
`content-generator`	Called by launcher	DeepSeek prompt chain adapts copy per channel (length, tone, hashtags)
`multi-publisher`	Cron hourly	Pulls from content_queue, publishes to the right API, marks published
`lead-processor`	Mautic webhook	UTM attribution, Listmonk list assignment, nurture kickoff, Chatwoot contact creation
`chatwoot-ai-agent`	Chatwoot webhook	Detects language, loads context, DeepSeek reply, escalates hot leads to Telegram
`analytics-loop`	Cron nightly	Umami event rollup, DeepSeek scoring, drafts new content ideas, notifies via Telegram

Content Queue Schema

Column	Type	Purpose
`campaign_id`	VARCHAR	Links every post back to the originating brief
`platform`	VARCHAR	pinterest / facebook / twitter / instagram / tiktok / email
`content`	JSONB	Platform-specific payload (copy, hashtags, media URLs)
`scheduled_at`	TIMESTAMP	When the publisher should fire
`status`	VARCHAR	pending / published / failed — drives retry logic
`response`	JSONB	Raw API response for audit + debugging

Adding a new channel is a matter of implementing one n8n HTTP node against the platform API and inserting rows into content_queue with the new platform value. The publisher does the rest.

Engineering Challenges Solved

📌 Pinterest Board ID Precision Loss

Pinterest Board IDs exceed Number.MAX_SAFE_INTEGER (>10¹⁸). Any dynamic JS computation silently produces wrong IDs and posts land on the wrong board.

Solution: Static string mapping in the publisher workflow instead of dynamic computation. Discovered after hours debugging "wrong board" errors — the kind of edge case you only catch in production.

🌐 Native Multilingual Copy (Not Translation)

Machine-translated marketing copy sounds robotic and kills engagement. But generating content 3× multiplies LLM cost and drift.

Solution: Single DeepSeek prompt with language-specific "voice" variables. Scripts are written natively in ES/EN/PT from the same brief, not translated. Cost stays flat, quality stays high.

⏱️ Rate Limits Across 6 Channels

Every platform has different rate limits: Twitter 1500/mo, Listmonk 300/day, Pinterest polite backoff, Facebook burst limits.

Solution: Centralized content_queue with per-platform scheduling windows. Redis tracks burst counters. Publisher respects platform caps and defers overflow to the next window automatically.

🔑 Instagram Meta Business Lockout

Meta restricted API access mid-project, killing direct publishing. A hard blocker for most stacks.

Solution: Pivoted Instagram to Chatwoot as the inbox layer, which preserves DM conversations and the AI chat agent. Publishing stays manual for IG, but capture + nurture kept working — loss contained.

📦 State Consistency Across 20+ Workflows

n8n workflows run independently. A lead can trigger lead-processor before chatwoot-ai-agent finishes writing the contact.

Solution: PostgreSQL is the single source of truth. Every workflow reads/writes through explicit upserts with conflict resolution. No in-workflow state, no race conditions.

🔊 AI Chat Without Hallucination

Letting a raw LLM answer lead DMs is a legal and brand risk. But scripted bots feel dead.

Solution: DeepSeek gets strict system prompt + contact context from Postgres + product catalog as retrieval. Hot leads detected by intent scoring are always routed to a human via Telegram before any commitment.

💡 This entire marketing OS pattern ports to any vertical. Change the brief schema, swap the language mix, point to different channel APIs — the orchestration layer stays the same.

Results & Metrics

Channels
orchestrated

20+

n8n workflows
in production

Pins ready
+ dynamic gen

Ad spend
CAC organic

Automation Coverage

Publishing Cadence

Pinterest3 pins/day/lang

Facebook1-3 posts/day

Twitter/X2-5 tweets/day (1500/mo cap)

Email (Listmonk)multi-list drip campaigns

Video (via Leon)3/day multi-platform

SaaS Replaced

Zapier / Make→ n8n

Mailchimp / SendGrid→ Listmonk

HubSpot / ActiveCampaign→ Mautic

Intercom / Drift→ Chatwoot

Google Analytics→ Umami

Monthly SaaS saved~$400+/mo

Deliverables

🤖

20+ n8n WorkflowsCampaign launcher, content generator, multi-publisher, lead processor, chat AI agent, intelligence monitor
📦

Content Queue SchemaPostgreSQL source of truth for every scheduled post, across every platform, with full audit trail
📧

Trilingual Email SystemListmonk (36 lists) + Mautic (9 campaigns, 15 emails, lead scoring) — all natively in ES/EN/PT
💬

Omnichannel Chat AgentChatwoot inbox with DeepSeek AI agent, language auto-detect, hot-lead escalation to Telegram
📈

Intelligence + AnalyticsUmami funnels + engagement scoring + A/B test feedback loop back into campaign planning
📚

Operations RunbookTelegram command interface (/campaign, /stats, /queue, /lead) and full docs for handover

Technology Tags

n8n PostgreSQL Redis DeepSeek GPT Image Listmonk Mautic Chatwoot Umami Pinterest API Facebook Graph API Twitter API TikTok API Telegram Bot Docker Oracle Cloud Multilingual Lead Scoring Self-Hosted

Need a Marketing OS That Runs Itself?

I build self-hosted, full-funnel marketing systems — from campaign brief to publish to capture to nurture. One VM, one codebase, zero SaaS lock-in.

💬 Hire Me on Contra ← Back to portfolio

What I Build

Voice AI Agents

Inbound receptionistsAny language

Outbound sales agentsCold + warm

Appointment bookingReal-time CRM

Emergency routingLive transfer

Content Automation

AI content generationMultilingual

Blog & social publishingMulti-platform

Email drip campaignsFull funnel

Trend intelligenceAutomated

Video Production

AI script writingNative multilingual

AI image generationGPT Image

TTS voiceoverWaveNet voices

Remotion renderCustom templates

Infrastructure

N8N workflowsSelf-hosted

PostgreSQL + RedisProduction-grade

Monitoring & alerts24/7

DocumentationRunbooks + guides

✅ Based in Honduras (UTC-6). Available for async collaboration across all timezones. Fluent in English, Spanish, and French.