Hi everyone,

I wanted to share a project I’ve been building in n8n — an AI-powered personal finance tracker that logs and analyzes expenses from natural language messages.

The idea was to make expense tracking as simple as sending a text message instead of filling out forms or spreadsheets.

What it does Users can send messages like:

“Lunch at cafe 250” “Amazon shoes 1200 delivered” “Paid Rahul 500” “Salary received 25000”

The system automatically:

Extracts structured data using an LLM

Classifies the transaction type (Online / Daily / Income / Expense)

Normalizes categories

Stores data in MongoDB

Sends a confirmation message

Generates weekly XLSX reports

Provides simple AI-based insights on spending

Core Features

1. AI Information Extraction An LLM node extracts:

Item

Amount

Category

Platform or store

Payment method

Notes

Order status

All from a single natural language message.

2. Transaction Classification Each message is classified into:

Online

Daily (offline)

Income

Expense (money sent to someone)

This classification controls routing and storage logic.

3. Category Normalization I implemented a basic “memory” system:

Categories are stored in MongoDB

The AI compares new categories with existing ones

Similar categories are reused

New ones are created only when needed

This keeps the data more consistent.

4. Multi-Path Logging There are different flows for:

Online purchases

Daily expenses

Income

Peer-to-peer transfers

Each is stored with slightly different fields where appropriate.

5. Weekly Reports A scheduled workflow:

Aggregates user data

Converts it to an XLSX file

Sends it to the user via Telegram

6. Conversational Finance Assistant There is also a simple conversational layer where users can ask for:

Spending summaries

Basic insights

Habit-related feedback

This uses short-term memory and stored transaction data.

Tech Stack n8n for automation

Telegram Bot API

LLM nodes for NLP and classification

MongoDB for storage

Redis for session memory

Airtable for long-term memory

XLSX generation for reports

Why I built it Mainly to:

Learn more about AI + automation design

Experiment with real-world LLM use cases

Build something practically useful

It started small and gradually evolved into a more complete system.

Looking for feedback I’d appreciate suggestions on:

Optimization

Data modeling improvements

Scaling AI-heavy workflows

Feature ideas