Damien Launay

Complete automation of incoming email management for a marketing agency facing saturation in its customer and sales services.

Design of a centralized system on n8n that analyzes, categorizes, and processes incoming emails using a RAG (Retrieval-Augmented Generation) architecture.

Semantic Analysis: Automatic classification of emails (Sales vs. Support) and sentiment analysis to prioritize urgent matters.

Dynamic Knowledge Base: Use of Supabase (Vector Store) and Redis to provide the AI with up-to-date factual context on offers and technical processes, eliminating hallucinations.

Remote Human Validation (HITL): Control interface via Telegram allowing teams to validate or correct AI-generated drafts before sending. Support for corrections via voice notes.

Bi-directional CRM Synchronization: Automatic enrichment of GoHighLevel contact records and logging of interactions.

Time Savings: Elimination of manual drafting for 80% of emails. Estimated savings of 25 hours/week for the team.

Speed: Response time divided by 10, increasing customer satisfaction and lead conversion.

Reliability: Standardized and accurate technical responses thanks to the AI's direct access to internal documentation.

Technical Stack

n8n, Python/JS, OpenAI (GPT-4), Anthropic, Supabase (Vector), Redis, GoHighLevel API, Telegram.

The agency wanted to scale its content production offering without increasing its staff. The manual process created a critical bottleneck.

The objective was to design a system capable of producing complete omnichannel campaigns (Blogs & Social Media) with almost no human intervention.

Solution:

Design of an advanced automation infrastructure on n8n. The architecture is modular, composed of a main orchestrator and several interconnected subsystems (over 130 logical nodes in total).

This system transforms a simple onboarding form into an executable content strategy.

Multi-LLM Orchestration: Intelligent routing between specialized models. Use of Gemini Vision for brand analysis (graphic charter extraction), Perplexity for factual research (anti-hallucination), and Claude Opus4 for complex writing.

"Self-Healing" System (Auto-correction): Development of an autonomous quality control loop. An "Auditor" agent assesses technical and semantic compliance, sending errors to a "Corrector" agent until strict validation.

Cross-Channel Production: Simultaneous generation of SEO articles (HTML with dynamic CSS injection), photorealistic visuals (via Flux Pro), and LinkedIn carousels.

Backend & Data: Management of persistence, history, and internal linking via Supabase (PostgreSQL) and real-time synchronization with the GoHighLevel CRM.

The deployment transformed the agency's operational model:

Productivity: Shift from 12 hours of human production to 8 minutes of machine time per campaign (-98% time reduction).

Profitability: Internal production cost divided by 20.

Scalability: Management capacity multiplied by 5 with the same number of staff thanks to code modularity.