Firsthand

Firsthand - See how it’s done, firsthand. * Background / Introduction Hands-on knowledge is hard to pass on. Experts rarely document every step, and learners have to pause and search through long videos while their hands are busy. Firsthand uses Muse Spark to turn first-person recordings into guides with short looping clips, clear instructions, and completion checks—ready to follow on an iPhone or Meta Ray-Ban Display glasses. * How it Works A creator records a task, narrates what they are doing, and can mark important moments. Muse Voice Transcribe provides a time-stamped transcript. Muse Spark 1.3 combines it with the first-person video to identify actions, set step boundaries, and write instructions and visible completion checks. Firsthand cuts a looping clip for each step, then lets the creator edit the draft before publishing. Learners can discover guides in the feed and move through them like a story. The iPhone app also includes a lens mode with tap and voice controls for Meta Ray-Ban Display glasses. The glasses capture and lens shown in our demo are simulated. * Tech We built an end-to-end SwiftUI app and Python backend around the models. To keep the guide aligned with the recording, we add a visible timecode to the video and validate step boundaries and quotes before showing them. The same pipeline created 47 guides from licensed first-person videos without per-video tuning. On 30 held-out videos with human-labeled steps, it beat an equal-length split baseline on 28 using a model-assisted step-matching metric. Firsthand makes practical knowledge easier to preserve, share, and follow while doing the work.