Mistral Chatbot Reply Scoring Workflow + Telegram Alerts (n8n)

Automatically score chatbot replies with Mistral—and alert your team in Telegram when things go wrong

The Mistral Chatbot Reply Scoring Workflow + Telegram Alerts (n8n) reviews every chatbot answer using Mistral Medium, stores an anonymized quality score in an n8n Data Table, and sends a Telegram alert whenever the reply is poor, the client seems unhappy, or a human handover is requested.

What this workflow does

• Starts on demand (manual execution with a sample conversation) or as a sub-workflow called by your chatbot.
• Merges incoming chat data with configuration (bot rules, Telegram chat ID, alert threshold, audit rate).
• Sends the conversation context to Mistral Medium, including the client message, the bot answer, and your rules—asking it to score accuracy, rule-following, tone, and whether a handover to a human is needed, plus the client’s mood.
• Parses the model’s JSON response, calculates an overall score, and prepares a short reason/excerpt for auditing (with explicit handling for model/parse errors).
• Writes one anonymized review row to the n8n Data Table (no chat identifiers).
• If flagged, sends a formatted Telegram alert including the score, mood, handover status, and the stored reason/excerpt.

Use cases

• Monitor chatbot performance in production by auditing reply quality and consistency with your bot rules.
• Trigger a human review when users appear unhappy or explicitly request support.
• Reduce support escalation time by routing only the problematic interactions to your team via Telegram.

Technical details

• Mistral Cloud credential for the Mistral Medium chat model.
• n8n Data Table named ai_reviews with fields: score, mood, status, reason, excerpt, lang, wants_human, handed_over, alert.
• Nodes used: Set, Code, Switch, Telegram, Data Table, Sticky Note.