
Millions of people get medical care in a country whose language isn't theirs. A lab report is then two barriers at once: medical jargon in a foreign language. What is off? How urgent is it? Which doctor should I see? I'm from Kyrgyzstan and study in Rome, so I know this problem first-hand. MedLens solves it in four steps: 1. READ: a vision model (DeepSeek-V4.1-Flash on AMD Token Factory) reads a photo or scanned PDF of the report in any language, for example Italian, and turns it into standard values. 2. CHECK: deterministic code compares every value with a reference table, so the AI never invents or changes a number. 3. EXPLAIN: a reasoning model (DeepSeek-V4-Flash on AMD Token Factory) explains the results in the user's own language. 24 languages are supported, including Russian, Kyrgyz, Chinese, Hindi, Urdu, Pashto, Arabic and Ukrainian, with right-to-left layout for Arabic, Urdu, Pashto and Persian. 4. ROUTE: the user gets an urgency level (routine / see a doctor soon / urgent), the specialist to visit, and practical questions to ask at the appointment. Safety by design: an urgency floor means the model can raise urgency but never lower it below what the code found; MedLens never diagnoses and uses cautious wording; the demo uses synthetic data only. MedLens is a navigator, not a diagnostician. Stack: AMD Token Factory (OpenAI-compatible API on AMD GPUs), Python, Gradio, Docker. Code: github.com/Askhat-Baialiev/medlens
18 Oct 2026