Keeping loans

Inspiration

Many lenders—especially community banks, credit unions and specialty servicers—lose loans to late detection, fragmented manual processes, and poor borrower engagement. I was inspired by the idea that timely, prioritized action (not just more data) is what prevents defaults. A desktop-first, offline-capable tool that blends deterministic rules with lightweight predictive signals can deliver that action where teams already work: on their desktops and in their servicing toolchains.

What it does

Continuously monitors payment history, covenants, and portfolio trends. Flags deteriorating loans with rule-based alerts and a simple ML risk score. Automatically generates remediation tasks, templated borrower communications, and assigns owners. Tracks documents, notes, and audit logs for every remediation action. Works offline (local encrypted DB) and optionally syncs to a hosted service for centralized reporting.

How we built it

Desktop client: Electron (or Tauri) + React + TypeScript for a responsive, desktop-first UI. Local storage: Encrypted SQLite (SQLCipher) for offline-first reliability and data security. Ingestion: CSV/Excel import and connector adapters for common core systems to accelerate onboarding.

Challenges we ran into

Data variability and access: core systems differ widely; field mappings and missing or noisy histories make scoring harder. Label scarcity: supervised models need labeled defaults/deteriorations; pilots often supply few examples.

Accomplishments that we're proud of

Converted the problem into a clear MVP that delivers measurable value in 30–60 days: import → alert �� task → resolution loop. Designed compact desktop wireframes and user flows focused on prioritized action and low-friction onboarding. Built a practical hybrid approach: conservative rule-based alerts first, with an easily interpretable ML score layered on top.

What we learned

Start with rules and human-in-the-loop workflows: they provide immediate value and data for later ML improvements. Explainability drives adoption: show why an alert fired and recommended actions, not just a numeric score.

What's next for Keeping loans

Build the clickable prototype and scaffold the desktop repo (Electron/Tauri + React) to validate UX with users.

Built With

• dockerfile
• html
• javascript
• typescript