Selected WorkProject 06AI Systems
RAG-PoweredClinic Assistant.
A retrieval-augmented assistant that answers clinic questions from a document knowledge base.
Clinic documents in Google Drive are embedded into a Supabase vector store, and an AI agent answers questions using that knowledge.
- n8n
- Supabase Vector Store
- Google Vertex AI
- Google Drive
06Project
of 15
- Clinic front desk
- Retrieval-augmented AI
- Questions answered from clinic documents
- Workflow built in n8n
01The Challenge
The same questions, again and again.
Clinic front desks answer the same questions about services and appointments again and again.
Answer from the clinic’s own documents, with a knowledge base that stays current.
The RAG workflow in n8n: the chat agent above, the Google Drive knowledge-base sync below.
02The Assistant
A chat agent and a self-updating knowledge base.
A retrieval-augmented assistant in n8n that answers from the clinic’s own documents, backed by a Supabase vector knowledge base synced from Google Drive.
- AI agent on a Google Vertex chat model
- Supabase vector store
- Watched Drive folder: new, updated and deleted files
03How It Works
Ask, retrieve, answer. Stay current.
Two branches in one workflow: answering, and keeping the knowledge current.
Ask
A chat message reaches the AI agent.
TriggerRetrieve
The agent searches the Supabase vector store for relevant passages.
RetrievalAnswer
It answers from the passages it retrieved, on a Google Vertex chat model.
AIStay current
New Drive files are embedded with Google Vertex and stored; updated files replace their entries; deleted files are removed.
Automatic
04Architecture
One knowledge base. Read on one side, kept current on the other.
The answering branch reads the vector store; the sync branch writes to it whenever the Drive folder changes.
01Capture
- AnsweringChat message
- SyncDrive folder: new, updated, deleted
02Process
- AI · Vertex chat modelAI agent
- Vertex embeddingsDownload and embed
03Knowledge
- Searchable knowledge baseSupabase vector store
04Deliver
- AnsweringAnswer from retrieved passages
- SyncKnowledge base kept current
- Workflow step
Answering and syncing run as two branches around the same Supabase vector store.
05Validation & Controls
Answers from the clinic’s own documents.
The agent answers from retrieved passages, and the knowledge base follows the source folder.
- Chat agent (Google Vertex)Built
- Vector store retrieval (Supabase)Built
- Drive sync: new, updated, deletedBuilt
- Only documents in the watched Google Drive folder are embedded.
- Updated files replace their earlier entries.
- Deleted files are removed from the knowledge base.
- The agent answers from passages retrieved from the vector store.
06The Stack
Four tools, one knowledge base.
Roles over logos. Only integrations that exist in the build are listed.
n8n
Chat agent and document-sync workflows
Google Vertex AI
Chat model and document embeddings
Supabase Vector Store
Searchable knowledge base
Google Drive
Source folder for clinic documents
07The Outcome
What the assistant does.
Repeat questions answered from the clinic’s own documents, with a knowledge base that updates as files are added, changed or removed.
Clinic documents in Google Drive are embedded into a Supabase vector store, and an AI agent answers questions using that knowledge.
- AI Workflow Implementation
- Knowledge Base Design
- Answers questions from the clinic’s own documents
- Retrieves relevant passages from a Supabase vector store
- Embeds new Drive files automatically
- Replaces updated files and removes deleted ones
Contact
Have a workflow, dashboard or CRM that needs fixing? Let’s talk.
Available for new projects
Tell me about the operation, the systems involved and where the work gets stuck. A short call is usually enough to see whether I can help.