Ghost in the Ledger: Open-Weight AI Finds Bank Leaks
A Nigerian developer creates Ghost in the Ledger for Hacktoberfest to detect hidden bank fees and spending ghosts using open-weight AI.
Stock photo for illustration only, not from the actual event
- Ghost in the Ledger is a submission for the DEV Hacktoberfest Open-Source AI Challenge Week 4
- Analyzes PDF statements, CSV files, or copied SMS alerts from global banks
- Uses deterministic code for exact figures and AI to rewrite them as spooky stories
- Supports fully local execution via Ollama or optional processing through Groq
A software developer based in Nigeria has created an open-source application called Ghost in the Ledger to help users identify hidden and forgotten expenses draining their bank accounts each month. Inspired by everyday banking alerts received via SMS or email—such as 2 AM POS transactions, minor text message fees, forgotten subscriptions, and salaries vanishing shortly after payday—the tool aims to solve the mystery of where money actually goes.
The application accepts bank statements in PDF or CSV formats, as well as pasted alerts from almost any bank and currency worldwide. It detects underlying financial patterns using precise code before allowing an artificial intelligence model to translate those verified facts into short, story-driven summaries that are both engaging and easy to audit.
Applying open-weight AI models to personal finance represents a significant milestone for data privacy, as users can run the entire workflow 100% locally on their own hardware using tools like Ollama. This approach ensures sensitive bank statement data never leaves the local machine, effectively eliminating privacy risks associated with cloud-based financial processing.
Stock photo for illustration only, not from the actual event
The platform identifies and categorizes several specific types of financial ghosts hidden within statements:
- Subscription ghosts: Recurring charges from the same merchant for roughly the same amount every month
- Fee poltergeists: Accumulated overdraft fees, ATM charges, foreign transaction fees, SMS costs, VAT, and stamp duties
- Midnight spenders: Debits occurring between 12 AM and 5 AM
- Payday vanishing acts: Large incoming credits where the majority of funds are depleted within three days
- Leaks: Numerous small debits directed to the same merchant
The system architecture is structured to prioritize accuracy and security. All ghost detection is handled by deterministic code, guaranteeing that identical inputs will always produce the exact same numbers and findings. The AI model is strictly restricted to rewriting these pre-calculated facts into narrative form and is never permitted to calculate or determine financial outcomes on its own.
"The Payday Vanishing Act. On 2026-08-31 a friendly ghost paid a credit of 2850 USD, then vanished after spending 2424.99 USD, leaving a spectral 85% balance."
gpt-oss-20b via Groq
Furthermore, the application incorporates a strict story guard mechanism to prevent models from inventing figures, names, or raw data strings. If any generated text fails validation checks—such as using spelled-out numbers, ambiguous quantities, or copying raw data rows—the system discards the text and replaces it with a standardized template built strictly from verified facts, ensuring users never encounter inaccurate financial figures.
Source: Dev.to
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment