AI fitness concierge & membership automation.
I built Forge Fitness as an end-to-end customer operations system: a customer-facing gym website backed by n8n workflows that connect conversation handling, business knowledge, vector retrieval, AI responses and operational routing.
From a gym website to a connected support system.
The project was designed to explore what happens when a customer-facing chatbot is treated as one layer of a larger operational system instead of a standalone AI widget.
Customers ask simple questions. The work behind the answers isn't always simple.
Membership questions, opening hours, class information and other routine requests can become repetitive manual support work. I used Forge Fitness to model a system that could handle those conversations while keeping the underlying business knowledge outside the frontend.
Separate the interface, orchestration, data and knowledge layers.
The website handles the customer experience. n8n orchestrates the workflow. Structured data can live in Supabase. Weaviate provides semantic retrieval. The AI Agent uses those components to interpret requests and produce grounded responses.
A typical customer request travels through a controlled path.
Context is part of the system.
The incoming payload is structured around conversation metadata such as conversation_id, message_id, timestamp, message role, message content and message history. That gives the workflow a foundation for handling multiple messages as part of one conversation.
Business knowledge can be ingested separately from the chatbot.
The supporting knowledge workflow uses a document pipeline that can move from a Google Drive trigger through file extraction and field preparation into embeddings and a Weaviate collection. This keeps knowledge maintenance separate from the customer interface.
The difficult part isn't “adding AI”. It's designing the boundaries around it.
This build reinforced the importance of validation, explicit data structures, retrieval quality, branching, testing and human handoff. The AI is one component inside the workflow rather than the workflow itself.
Read the full build breakdown
Architecture responsibilities
- Frontend: provides the Forge Fitness website and embedded concierge experience.
- Webhook layer: receives structured customer messages and metadata.
- Data handling: normalizes and validates incoming payloads before downstream processing.
- Conversation layer: works with conversation and message identifiers plus message history.
- AI layer: uses an AI Agent to interpret requests and coordinate tools.
- Retrieval layer: uses Weaviate to retrieve relevant documents for grounded responses.
- Persistence: structured records can be stored and retrieved through Supabase.
- Routing: branches determine the response path and provide room for human assistance.
- Knowledge ingestion: source files can be transformed into embeddings and added to the vector collection.
What this demonstrates
This project demonstrates practical experience connecting a frontend interface to an n8n automation backend, handling structured conversation data, connecting an AI Agent to a vector store, and designing an automation around real customer-support scenarios.
Production considerations
Before using a similar architecture for a real business, I would further harden authentication, rate limiting, observability, failure handling, AI evaluation, retrieval evaluation, data retention and escalation tooling.