Samuel Halimi@sam_halimi

NOVA turns a folder of contradicting project documents into one screen: an answer, a proof that opens at the exact passage, settled contradictions and actions with owners. An AI assistant updates it from an email, never without approval. Offline.

After a sexual assault, the first hours matter and the path is confusing. People do not know where to go, drug traces fade from blood within about a day and from urine within about five, and survivors retell their story to one professional after another while care teams rebuild a timeline from scattered notes. Boussole is an installable, mobile-first web app in French and English, with two sides. For the person, it shows help first (the 24/7 support line and 911), then the official care steps in order with their deadlines and sources, a map of verified Montréal resources, a private story encrypted on the device, and, only if they decide, a file prepared for a free legal-advice service with double consent. No AI touches this side, and the story never leaves the phone. For the care team at a designated centre, Boussole offers per-item consent, a rule engine that sorts forensic samples by time window, and an AI-assisted timeline. The AI is the open model DeepSeek-V4-Flash, served by AMD Token Factory on AMD GPUs through an OpenAI-compatible API, called server-side only. Notes are pseudonymised in the browser and checked again on the server, which refuses to send anything that still contains an identifier. A line is kept only if it quotes the notes verbatim; culprit, credibility, score or diagnosis wording is removed and shown; a human validates every line, and each action is sealed in a SHA-256 hash-chained log. On fictional notes, a timeline typically takes 35 to 60 seconds on this trial endpoint, and every line kept by the guardrails quotes the notes verbatim. Uncertainty stays visible: the live answer is labelled with the provider and model, and if the AI is unavailable a pre-recorded answer is shown and labelled as such, with the reason. The demo uses fictional data only. The AI guides, the human decides.

Six Matches. One leaderboard. Build on AMD, earn points and compete for prizes.

🌎 Hybrid Hackathon | 💻 Online Build: 12-17 October 2026 | On-site phase: 17-18 October 2026 To participate in the hackathon, simply click Sign up with AMD. If you are not already a member of the AMD AI Developer Program (ADP), you will need to create an account. The AMD Developer Hackathon: ACT III is a hands-on event for developers, founders, engineers and builders who want to push what’s possible with AI on real infrastructure. 📍 If approved for the on-site experience, you'll also have the opportunity to showcase your project live. Travel and accommodation will not be covered.