Personal AI Bot

Help me Build my Workflow
1

Hey n8n Experts!!

I am new to n8n. I want to build an ai bot having knowledge base of youtube video (hindi and english videos) and conversation must be remember by ai and answer with previous conversation and other knowledgebase be same

I hope I clearly mention my requirements.

1 Like
2

Hey @Digispyzone, welcome to n8n!

What you’re describing is a RAG (Retrieval-Augmented Generation) setup combined with conversation memory — fully doable in n8n. Here’s the architecture I’d recommend:

1. YouTube Knowledge Base
Fetch transcripts via a service like Supadata or AssemblyAI (they handle both Hindi and English well). Then chunk the transcripts, embed them using a multilingual model like multilingual-e5-large, and store them in a vector database — Qdrant, Pinecone, and Supabase pgvector all integrate natively with n8n’s vector store nodes.

2. Conversation Memory
The AI Agent node has built-in memory options. “Window Buffer Memory” covers short-term context within a session. For cross-session memory (bot remembers past conversations), store chat history in Postgres or SQLite and inject it into the system prompt on each new message.

3. Putting It Together

  • AI Agent node → Vector Store Retriever (YouTube knowledge) + Memory node
  • On each user message: retrieve relevant transcript chunks + recent conversation history → send to LLM → return answer

The trickiest part is usually preprocessing the transcripts — video transcripts often lack punctuation, so you’ll want to clean them up before chunking to get better retrieval results.

We’ve built similar pipelines at Noyra-X. Happy to help if you get stuck on a specific step!

3

Hi @Digispyzone Welcome!
AI Automation Specialist here, let me help you, first get yourself bunch of youtube video transcriptions (yes transcriptions, cause we need the AI to understand the context), once you have those, goto Pincone, create a vector database and now everything is good to go, just hop into n8n setup a RAG workflow you will see on google you will get it, inject all those transcribes to pinecone store, and now once that is done you can move ahead to creating your chatbot by adding a chat trigger attached to AI agent and a simple memory + an OpenAI model and in tool attach pinecone vector store and that is really it, let me know if you get stuck somewhere!