HOLOFIXER

HOLOFIX: see inside any device, repair what you can. Most broken laptops are thrown away because their owners don’t know what is wrong, whether it can be fixed, or what it would cost. Right-to-repair laws only help if repairing is understandable. HOLOFIX makes it visual and honest. Type any device (“ROG Strix G16”, “MacBook Air M1”, even a car like a Renault Clio IV): Identify: Muse Spark searches the web for every model sold under that name, asks which one you have, then finds its possible configurations. Parts that are soldered and can never be upgraded are marked. Explore: it reads cited teardowns and spec sheets and extracts 15–30 parts with positions, connections and a source for each. The browser builds the device as a 3D hologram from that data, in code, with no hand-made models. You can rotate it, assemble it with a slider and explode it again. Click any part to see what it does, its spec and whether you can replace it. Diagnose: describe the problem in one sentence or add a photo. The likely parts light up on the hologram, ranked, with reasons. HOLOFIX asks the one question that best narrows them down, and fixes are tiered green / yellow / red. Repair: one click writes a printable repair guide with tools, cited steps, and real prices for doing it yourself, using a repair shop or buying new. In the demo, a dead trackpad costs about €178 to repair versus €2,597+ for a new laptop. Trust is enforced in code, not in prompts: Every source shown must be a URL the search actually returned; invented links are deleted. Screw specs, prices and steps are kept only when cited; totals are computed by code. Dangerous cases (swollen batteries, smoke, mains voltage, airbags, brakes, EV high voltage) are a hard stop: diagnosis only, “see a professional”, no steps. Every model answer is validated against a schema; suspects must exist in the device’s parts list. If extraction fails, a generic layout is shown and labelled “Unverified”. Evaluation: on 32 documented symptom-to-cause cases, the right cause is in the top 3 in 97% of them, and 5 of 5 dangerous cases are stopped. These are common, documented failures written by me, so unusual real-world faults will score lower; the full results are in the repo. Stack: Python (FastAPI) backend, plain HTML + Three.js front end, no build step. Devices and diagnoses are saved in the browser.