Gemma 4 reads, code decides: offline NDIS invoicing that caught 20 of 22 restraints

Description and use

Arvo leverages a local Gemma 4 model to convert handwritten or photographed shift diaries into compliant NDIS invoices and restrictive‑practice alerts, all offline.

Problem

Independent NDIS support workers must convert handwritten or photographed shift diaries into compliant invoices and flag restrictive practices, but existing tools are expensive or require online services.

Outcome

In a test week with 7 shifts, Arvo produced 10 invoice lines, 3 red‑flag panels, and 13.3 s processing time on a laptop RTX 4050. It correctly flagged 20 of 22 restraints in the diary.

Workflow

• 1. Worker inputs diary by typing, photographing, or dictating.
• 2. Gemma 4 (Ollama) reads the input and extracts shifts and notes.
• 3. Code validates dates, times, participant names, and calculates hours.
• 4. Pricing rules apply: NDIS pricing schedule, Victorian public‑holiday calendar, and schedule s2.4(c).
• 5. Code identifies restrictive‑practice notes, quotes the sentence, and assigns notify‑by dates.
• 6. Outputs invoice lines, printable invoice, CSV, and a red‑flag panel.
• 7. Worker reviews and finalizes before sending to plan manager.

Evidence

Gemma 4 did far worse than on the rendered test photos: 0 of 7 shifts fully right.

Source

The page is docs/demo/run.json rendered: gemma4:e4b on docs/demo/diary.txt, 7 shifts, 10 invoice lines, 3 flags, 13.3 seconds on a laptop RTX 4050 (6 GB).

Source