Lognos

We're building Safety Advisor to help safety technicians do their job and to help save lives on construction sites. This isn't a demo built for judging: we're working to put it into real use. A safety technician covers several crews and sites, and a large part of the day goes to reading work plans, checking them against the regulation and writing up what they see on site. Safety Advisor takes on that load. Crews send photos, video and voice notes from site through Telegram or the web app. The advisor ties each one to the job's safety plan, cites the regulation it relies on (an Ontario O. Reg. 213/91 extract) and says where evidence is missing. The technician stays in charge: the output is labelled as an adviser assessment and never counts as formal safety approval. What existed before the event: a private Safety Advisor I built earlier (chat workspace, incident recall, work-plan extraction, recommendations, canvas and dashboards). That code is not in the submitted repository. Built during the event (Oct 31 – Nov 4): - Meta model integration: Muse Spark Open on Baseten and Nebius, Muse Spark 1.3 and Muse Wonder 1.0 through the Meta Model API, and SAM 3.1. - Photo privacy and assessment: faces, licence plates and company marks are redacted with SAM 3.1 before anything is stored or returned. - Video privacy pipeline on a durable private worker with checkpoints and recovery. - Voice notes processed with Muse Wonder. - Telegram field channel linked to saved jobs. - Saved jobs with a safety plan built from the work package, the company standard and the regulation extract. - Private per-user workspace on Supabase Postgres with row-level security, retention and purge. - Spend limits per run, per day and overall. - A synthetic corpus: all demo data is fictional. The README lists every new feature with its files.