AI-Powered Customer Support Assistant

I built an AI assistant that automates customer support replies inside Gorgias, helping teams respond faster while maintaining a consistent brand voice. The assistant uses real-time ticket data, tone training, and internal documents to generate high-quality, on-brand replies with minimal human intervention.

✅ Reduced manual workload for agents

✅ Maintained brand tone across all replies

✅ Improved response time and accuracy through automation

Support teams often struggle to write replies that are fast, accurate, and aligned with the company’s voice, especially when relying only on previous examples. This leads to inconsistency, slower replies, and higher agent fatigue.

I created a system that combines AI, workflow automation, and document-based context (RAG) to generate draft replies inside Gorgias.

🧠 RAG-Enhanced AI Replies

Admins upload brand documents via the frontend (Lovable)

I store these in Supabase Storage and process them with n8n

Text is extracted, chunked, embedded, and saved using pgvector

During ticket reply generation, I fetch relevant chunks for prompt context

🎯 Tone Training

I added the ability to upload tone/style examples

Admins can select a tone like “Friendly” or “Professional”

The system dynamically injects tone guidance into the AI prompt

🔐 PII Anonymization

Before generating replies, I strip out all personal/sensitive data to ensure compliance and data safety

📝 Ticket Reply Logging

Every draft response and action is logged to Supabase for full visibility and QA

n8n – Manages workflows, webhook triggers, and AI pipelines

Supabase – Stores documents, prompts, logs, and ticket metadata

pgvector – Enables document search and semantic matching

OpenAI – LLMs used for generating human-like replies

Gorgias – Customer support platform for ticket management

React – Build a dynamic, responsive admin dashboard with modular, reusable components for enhanced interactivity and flexibility.

Tailwind CSS – Used for building responsive, clean UI components