Managing rental properties in emerging markets like Nigeria is often a logistical nightmare, relying on faded thermal paper receipts and unindexed WhatsApp screenshots. A new open-source project, SlipsLedger, aims to fix this by using Google's Gemma 4 (gemma-4-26b-a4b-it) to extract structured data from payment slips. Submitted for the Hacktoberfest Open-Source AI Challenge, the tool converts chaotic photo trails into audit-ready financial ledgers, reducing the hours caretakers spend reconciling payments.

The Problem With Informal Payments

In many developing regions, rent, security dues, and utility fees are settled via mobile bank transfers like Moniepoint, OPay, or Kuda. Tenants typically send a screenshot of the transaction or a photo of a paper slip to a landlord's WhatsApp. This workflow is fragile; thermal paper fades, rubber stamps blur, and critical details like dates are often cropped out. When disputes arise, landlords and tenants lose days scrolling through gigabytes of images, leading to accusations of non-payment and even police mediation.

How Gemma 4 Powers the Extraction

SlipsLedger bypasses brittle OCR pipelines by relying directly on Gemma 4's multimodal vision capabilities. The application accepts PNG, JPG, or WEBP images and uses the google-genai SDK to extract key fields such as payment type, amount, currency, date, and transaction references. The system enforces a strict JSON schema contract to prevent hallucinations, ensuring that every extraction includes specific keys like risk_flags and whatsapp_draft. This approach allows the model to handle skewed angles, glare, and diverse banking interfaces without fallback to closed models.

Local-First Architecture and Privacy

The tool is built as a local-first web application using Streamlit for the UI and SQLite for persistence. Images are downscaled in memory to a maximum longest edge of 1600 pixels to protect host RAM, keeping consumption under 150MB. Crucially, the architecture prioritizes data sovereignty. Because Gemma is open-weight, it can be served locally or on a private cloud instance, ensuring that sensitive personally identifiable information (PII) like bank account numbers and addresses never leave the operator's trust boundary.

Real-World Validation in Ibadan

To test the tool's viability, the developer conducted a live trial with Mr. Kolawole, an estate caretaker managing a 12-unit block in Bodija, Ibadan. They processed eight real payment screenshots, including electricity bills and security dues. Gemma 4 accurately extracted amounts and references, and correctly categorized a payment captioned simply as 'October cleaning money.' The system flagged a cropped timestamp on one OPay slip, a common tactic used by tenants to reuse old transfers. Mr. Kolawole offered N3,000 ($2.00) monthly for the service, validating the demand for affordable, focused tooling.

Key Takeaways

  • SlipsLedger uses Gemma 4 (gemma-4-26b-a4b-it) for multimodal extraction, avoiding complex OCR stacks.
  • The tool enforces strict JSON schemas to ensure data integrity and detect dispute risks like missing dates.
  • Local-first architecture with SQLite ensures financial data privacy and reduces costs compared to closed APIs.
  • Real-world testing with an estate caretaker in Nigeria validated the product-market fit for low-cost property management tools.

The Bottom Line

SlipsLedger proves that open-weight models can outperform expensive, opaque SaaS solutions in niche, high-friction markets. By prioritizing data sovereignty and cost-efficiency, it offers a scalable blueprint for building AI tools that serve the economic realities of emerging markets.