Rangga Kori Lesmana

AIRA is a Windows desktop AI co-worker for a clinical diagnostics lab, running entirely on Meta Muse: Wonder for answers and tools, Spark for quick decisions, Voice Transcribe for listening, and Image for visuals. Its voice comes from ElevenLabs. The problem: every month, lab staff reconcile reagent usage sheets against physical stock counts by hand. It takes a day, the numbers rarely match, and nobody can tell why. Generic chatbots make this worse, because they invent numbers. What we built during the event, on top of an existing internal assistant (baseline commit a693286): Voice co-worker. AIRA greets you by name and reads a morning briefing: sales up or down, rising and falling tests, what to order, a suggestion, and a health trend from the web. Then you just talk. Muse Spark turns each spoken request into one safe, permission-checked action, and AIRA opens the right screen by itself. Reagent Check. Reconciles usage sheets and stock counts against the tests the lab really ran. It flags wrong item codes, inflated per-test factors, over-consumption and unit mismatches, and every finding links to its exact file › sheet › row. Reorder & voice recount. A reorder list computed by code in whole packs, an AI summary, Excel export, and a shelf recount done out loud. Number check. Every number Muse writes is checked against the facts the code computed. If one doesn’t match, it is marked, or the briefing falls back to a version written by code. Trust by design: numbers come from code, not the model. Patient data never leaves aggregate tools. All demo data is synthetic.