Mrinal Desai

FaultLens is a diagnostic assistant for field technicians. A technician on site has to decide what is wrong from what they can see, hear and measure, usually alone, so they guess: good parts get replaced, customers get repeat visits, and live equipment gets opened. FaultLens takes the evidence a technician can collect (the sound of the unit, thermal images, a sensor log, photos of parts, the unit's fault history, old handwritten job cards and spoken notes in Hindi or English) and ranks the likely causes with the evidence for and against each. It gives the one check to do next, tells a faulty part from one that only shows a symptom, says how far damage has gone and how long a part has left, and says "not certain yet" when the evidence is thin instead of guessing. Fixed safety rules block any step that needs the power isolated, and a second model reviews every diagnosis. Meta models: Muse Spark reasons and coaches, SAM 3.1 finds parts in photos, Muse Voice Transcribe takes spoken notes, and Muse Spark Open (Baseten, Fireworks) checks each call and reads handwriting. The worked example is the split air conditioner. On 14 test cases it was right 13 times, against 7 for simple sensor rules, and right every time it said it was confident. Everything was built during the hackathon. Live: https://mrinal-datsci--faultlens-web.us-east.modal.run