🧠 System Architecture

This assistant is built on top of n8n, using LangChain, Mistral Cloud, and Google Sheets to enable a natural-language-based finance tracker.

📐 Components

Component	Purpose
Chat Trigger	Receives user messages
LangChain Agent	Parses intent and picks a tool
Mistral Cloud	Powers the LLM responses
Google Sheets	Stores all persistent expense data
Calculator Tool	Computes totals and averages
Chat Memory	Stores short-term chat history

🔄 Workflow Diagram

🧠 AI Prompt Strategy

The AI is instructed with a strict system prompt that:

• Describes the exact schema (Date, Details, Category, Amount)
• Lists accepted categories
• Forces confirmation before entry
• Bans hallucinated or assumed data
• Uses Calculator only for math

⚙️ Tool Descriptions

Each tool node has a toolDescription configured like:

• Add Expense: Append a row to sheet after confirmation
• Get Expense By Category: Return rows matching a category
• Get All Expense: Return all rows
• Calculator: Only used when asked to compute totals/averages