Ledgerly: A Private, Local-First Receipt and Invoice Assistant

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle

What I Built
Ledgerly is a local-first receipt and invoice assistant for freelancers, small businesses, and anyone who wants a clearer view of their spending without sending financial documents to a cloud AI service.

Users can upload a receipt or invoice as a JPG, PNG, or PDF. Ledgerly extracts the document text in the browser, sends only that extracted text to a locally running open-weight Phi-3.5 Mini model, and converts the result into structured financial fields such as:

Vendor or company name
Invoice and transaction dates
Customer and billing details
Line items
Quantities and prices
Subtotal
Tax
Total amount
Payment method and notes
The extracted information is then presented in a modern finance dashboard for quick review and organization.

The people most likely to pay for Ledgerly are freelancers, independent contractors, small-business owners, and bookkeepers who need lightweight expense tracking but do not want to manually enter every receipt or upload sensitive documents to a third-party AI API.

Demo
Live demo:

A useful demo flow is:

Open Ledgerly.
Choose the built-in sample invoice or upload a receipt image.
Let the browser perform OCR or PDF text extraction.
Run local inference with Phi-3.5 Mini.
Review the structured invoice fields in the dashboard.
Code
Source code:

How I Built It
Ledgerly is built with Next.js and React. The interface uses a dashboard-style design inspired by modern consumer finance and social applications, with strong visual hierarchy, cards, activity views, and responsive layouts.

The AI pipeline is intentionally client-side:

The user selects a JPG, PNG, or PDF.
The file is processed in the browser.
OCR or PDF text extraction produces plain text.
The extracted text is passed to Phi-3.5 Mini through WebLLM.
The model returns structured invoice data.
Ledgerly validates and displays the extracted fields for review.
The original document is not uploaded to a remote AI endpoint. This keeps the prototype privacy-focused and demonstrates that useful document intelligence can run directly on a user's device.

The app also includes a built-in sample invoice so the complete workflow can be demonstrated without requiring users to find a test document online.

Why Does Open Innovation Matter?
Open innovation made it possible to build Ledgerly around a model that users can inspect, run, and control instead of depending entirely on a closed AI API.

A closed API would make the initial implementation easier, but it would require sending financial documents or their contents to an external provider. It could also introduce usage costs, vendor lock-in, rate limits, and uncertainty about how sensitive data is handled.

Using an open-weight model with WebLLM provides several important advantages:

Private, browser-based inference
No per-document AI API fee
Reduced dependence on a single provider
A transparent and reproducible technical stack
The ability to improve prompts and parsing logic independently
A practical demonstration of local AI for everyday business workflows
There are trade-offs. Browser inference requires a compatible device, model download, memory, and processing time. However, for a lightweight receipt assistant, the privacy and ownership benefits make local inference a compelling option.

Prize Categories
Ledgerly is being submitted for categories related to:

Open-source AI
Open-weight models
Local inference
AI-powered productivity tools
Privacy-focused applications
What I Would Build Next
The next version could add correction feedback, CSV and accounting exports, recurring vendor detection, receipt search, tax-category suggestions, and optional encrypted synchronization between devices.

The long-term goal is to make Ledgerly a practical personal finance tool where AI is helpful without requiring users to give up control of their financial documents.