02Advanced n8n - Telegram
Autonomous Support Agent
A Telegram agent that classifies intent with an LLM, answers from a RAG knowledge base, and escalates sensitive cases to a human.
n8nTelegramOpenAIRAG Engine APIQdrant
The problem
Support inboxes mix repetitive questions with things that genuinely need a human (billing, complaints). Everything should be triaged automatically before a person looks at it.
What it proves
Handling real autonomous business flows safely: grounded answers, conservative escalation, and production error handling.
Architecture
- 1.Telegram Trigger receives each message
- 2.LLM classifies intent into a fixed taxonomy (JSON mode)
- 3.A Switch routes: questions go to the RAG knowledge base, sensitive cases escalate
- 4.If the RAG answer isn't confident, the agent escalates instead of guessing
- 5.Error Trigger workflow alerts the team if anything fails after retries
Architecture diagram
Engineering highlights
- +Conservative escalation: never guesses on billing/complaints
- +Grounded answers only (answered=false -> escalate)
- +Versioned intent-classification prompt + full error handling
Demo
Demo: question answered from KB; refund request escalated to the team.
Demo recording placeholder (add a Loom link or demo.gif)