FamilyVault: An Offline-First AI Document Safe Built for My Parents

Description and use

FamilyVault enabled parents to locate documents quickly, detect date‑of‑birth discrepancies, and answer policy questions with cited evidence—all while keeping data offline and encrypted.

Problem

Parents needed a secure way to store and search family documents without uploading sensitive data to cloud services.

Outcome

FamilyVault enabled parents to locate documents quickly, detect date‑of‑birth discrepancies, and answer policy questions with cited evidence—all while keeping data offline and encrypted.

Workflow

• 1. Drag‑and‑drop PDFs or images into FamilyVault. 2. Local OCR (Tesseract.js) extracts text. 3. Local Gemma‑4 model classifies document type and proposes metadata. 4. Extracted facts are validated with regex/schema and stored in SQLite/SQLCipher. 5. Contradiction engine cross‑checks facts across profiles and flags inconsistencies. 6. Users query via natural language; Gemma‑4 generates answers citing document lines. 7. All data remains encrypted on disk (AES‑256‑GCM) and never leaves the machine.