FinDroid – Automated UPI Transaction Extractor

FinDroid is an AI-powered Android automation system that extracts daily UPI transactions from popular payment applications and stores them in a single, continuously updated CSV file.

It eliminates the manual, repetitive, and error-prone process of checking transaction histories across multiple apps—making it ideal for merchants and high-frequency UPI users.

---

Table of Contents

• Introduction
• Problem Statement
• What FinDroid Does
• System Architecture
• Tech Stack
• LLMs Used
• Project Structure
• Setup Instructions
• How to Run FinDroid
• Output Format
• Security and Privacy
• Use Cases
• Known Limitations
• Future Enhancements
• Author

---

Introduction

Merchant users handle dozens or hundreds of UPI transactions daily across multiple payment apps.
FinDroid automates transaction extraction using LLM-driven Android UI reasoning, consolidating all transactions into a reusable CSV file.

---

Problem Statement

Manually opening each payment application, navigating transaction histories, and maintaining financial records is:

• Time-consuming
• Inefficient
• Error-prone

FinDroid automates this workflow to reduce manual effort and improve accuracy.

---

What FinDroid Does

• Automates Android UI interactions using LLM-driven reasoning
• Extracts transactions from:
  • PhonePe
  • Google Pay
  • Paytm
• Accepts interactive user input:
  • Target date (default: today)
  • Target application
• Saves all transactions into one CSV file
• Appends data if the file exists, otherwise creates a new file
• Avoids brittle UI scripts using agent-based navigation

---

System Architecture

User Input (Date + App)
        ↓
Prompt Instructions (prompt.py)
        ↓
DroidRun Agent
        ↓
LLM-Guided Android UI Automation
        ↓
Transaction Extraction
        ↓
Single CSV File (Create / Append)

---

Tech Stack

Core Technologies

• Python 3.10+
• DroidRun – Android UI automation framework
• OpenRouter – LLM gateway
• Colorama – CLI styling and banners

---

LLMs Used

• mistralai/devstral-2512:free (via OpenRouter)

The LLM is responsible for:

• UI understanding
• Navigation and decision-making
• Robust interaction with dynamic Android interfaces

---

Project Structure

findroid/
│
├── main.py
├── prompt.py
├── all_upi_transactions.csv
│
├── .venv/
└── README.md

---

Setup Instructions

Clone Repository

git clone https://github.com/yourusername/findroid.git
cd findroid

Install & Run

python -m venv .venv
source .venv/bin/activate

pip install droidrun llama-index colorama python-dotenv
python main.py

---

Output Format

merchant_name,datetime,amount,direction,app
Amazon,2026-01-16 10:32,499,debit,googlepay

---

Security

• No credentials hardcoded
• Local-only processing
• Secure environment variables

---

Author

FinDroid was built to automate financial record-keeping using LLM-driven Android UI automation.

• Rohit Kumar
• Swetankan Kumar Sinha
• Harsh Kartik Singh