Dindo — AI health-case manager on AMD GPUs

In July 2022 a car hit me. Surgeries, physiotherapy, a post-thrombotic syndrome and a lawsuit. The hardest part for my family was the paperwork: handwritten prescriptions from five doctors, almost 28,000 Bs in receipts, invoices the insurer refused, even studies ordered for the wrong knee. Clinics have systems. The family has a bag of papers. Dindo turns that bag into an organized, audited case. Take a photo of any paper: prescription, invoice, order, report, court summons. Gemma-4-31B-IT, an open vision model on AMD GPUs via AMD Token Factory, reads it, even handwriting or sideways photos, and Dindo files everything: - Today: 24-hour dose grid, overdue doses, reminders. - Calendar, instructions (DO / AVOID), diagnoses as written, test orders. - Expenses and insurance: who paid, the insurer's deadline rule, claim status and an AI-drafted claim PDF. - Ask your case: DeepSeek-V4-Flash on AMD answers, linking every fact to its document. Dindo audits the case with AI plus rules. In my own papers it found X-rays ordered for the right knee when the fracture is the left one, an invoice dated before the sessions it charges, and 4,660 Bs of invoices past the insurer's deadline. It never guesses: if a drug name or body side is illegible, it asks the family. It never diagnoses. The demo is my real case with personal data covered before any AI call. Gemma read all 166 documents on AMD; on a benchmark it gets 98% of fields right and catches 5 of 5 planted errors. Every AMD call is visible in the app. With open models, the same app can run on a clinic's own AMD hardware, so health data never leaves it.