Support request - FAQs and Calendar Chatbot

[shanik]

Hi, I’m creating a flow that will function as a chatbot (in this case, the example is my own automation business), but there’s so much information that I’m a bit lost.
I have a webhook coming from ManyChat, a buffer, a split for text or voice, and three subflows.
I have a FAQ subflow, and this is where I’m stuck. I have an AI agent using Google Sheets, but I recently came across the concept of RAGs (Real-Time Aggregate Groups). I’ve spent many hours searching for information, and sometimes I find that the architecture I’ve been using is already obsolete or that there are tools that make certain processes much easier.

Any ideas or improvements for the overall workflow? What tools are currently used for this type of workflow?

How can I improve the accuracy of the FAQ flow? The same goes for the calendar; it schedules, but sometimes it gets the right day, sometimes it puts it on a different day.

[achamm]

You’ve got a complex workflow, but let’s focus on improving accuracy. For the FAQ, consider using the “OpenAI” node with a “Text Embeddings” model to create embeddings from your FAQ data. Then, when a user asks a question, embed their query and compare it to your FAQ embeddings to find the most relevant answers.

Regarding the calendar, double-check how you’re parsing the date information from the user’s input. Ensure that the date format is consistent and that the calendar node is correctly configured to interpret it. You might need to use a “Set” node to format the date before sending it to the calendar.

[Miliaga]

@shanik First, RAG stands for Retrieval-Augmented Generation, not “Real-Time Aggregate Groups”. It’s a technique where:

1. Your knowledge base is stored in a vector database

2. User questions are converted to embeddings (numerical representations)

3. Similar content is retrieved from the vector database

4. That context is fed to the AI to generate accurate answers

Why RAG is better than Google Sheets for FAQs:

• Handles large amounts of unstructured data

• Finds semantically similar content (not just keyword matching)

• More accurate responses

• Scales better as your knowledge base grows

For FAQ/Knowledge Base:

Vector Store (RAG) - Recommended

• Use n8n’s Pinecone, Qdrant, or Supabase Vector Store nodes

• Store your FAQ content as embeddings

• Use Vector Store Retriever to find relevant answers

• Feed results to your AI Agent

For Calendar Booking:

Your date accuracy issues likely come from:

• Ambiguous date references (“next Tuesday”, “this Friday”)

• Timezone confusion

• AI not having current date context

Improvements:

1. Always provide current date/time to the AI in system prompt

2. Use structured output to extract date components

3. Add validation before booking

4. Use a dedicated calendar tool in your AI Agent

[Miliaga]

For FAQ with Vector Store:

Setup:

1. Create a Document Loader workflow:

   • Load your FAQ content (from Google Docs, Notion, or files)

   • Use Default Data Loader node

   • Connect to Pinecone Vector Store or Qdrant

   • Run once to populate your vector database

2. In your main workflow:

   • Use AI Agent node

   • Add Vector Store Tool

   • Configure it to search your vector database

   • The agent will automatically search when users ask questions

For Calendar Accuracy:

Add this to your AI Agent system prompt:

Current date and time: {{$now.format('YYYY-MM-DD HH:mm')}}
Current day of week: {{$now.format('dddd')}}

When booking appointments:
1. Always confirm the exact date with the user
2. Use ISO format (YYYY-MM-DD) for dates
3. Ask for clarification if the date is ambiguous
4. Repeat the date back to the user before booking

Create structured calendar tools:

Tool 1: Check Availability

• Input: date (YYYY-MM-DD), time preference

• Output: Available time slots

• Use HTTP Request to your calendar API

Tool 2: Book Appointment

• Input: date, time, customer name, service type

• Validation: Check if date is in the future

• Output: Confirmation with booking details

This should help!