Project Story

Inspiration

Anyone who has walked through an apartment before moving in, checked contractor work during a remodel, inspected a property, or photographed cables before unplugging a desk knows the same annoyance. Stopping after every shot to unlock your phone and type notes with your thumbs turns a quick five-minute walk into a chore, while taking fifty unnamed IMG_4021.JPG photos leaves you squinting at drywall and wires for hours that evening trying to remember which room was which.

Apps that turn a walkthrough into a punchlist, such as MakePunchList ($19.99 to $29 per report) or CompanyCam ($79+ per month), charge recurring fees, upload private home or client rooms to outside servers, and cannot pair photos from a separate camera with a voice recording. Since my phone videos, photos, and voice memos already back up to my NAS at home, I wanted the NAS itself to turn a natural talking walk into labeled photos and a finished checklist privately, at zero per-report cost.

What it does

SyncWalk turns a hands-free walkthrough into clearly named photos and an organized Markdown checklist right on a UGREEN AI NAS.

You can capture a walk in two natural ways:

  1. Film one talking video as you walk (split locally into its audio track and still frames, for example via a local ffmpeg container in UGOS Docker), so you just point and talk with one hand.
  2. Snap photos while a voice memo runs (on a phone or a separate camera/drone/thermal imager), which runs directly with zero extra containers.

When you ask Uliya to compile the folder, SyncWalk transcribes your recording on the NAS and lines up the exact seconds you spoke each comment with the matching video frames or photo capture times (skipping empty hallway frames in between). If you used a separate camera whose clock or timezone did not match your voice recorder, SyncWalk checks one or two recognizable shots (such as a kitchen sink or a water heater label) to work out the time difference automatically before matching the rest of the walk. It then uses on-device image and text recognition on the selected shots (up to 30 per run) to check visible details and read serial numbers off equipment stickers.

Real walks are messy, so SyncWalk never forces fake one-to-one matches. If you say a reminder out loud without pointing at anything ("order two outlets for the garage"), it keeps that note in its own action-items list. If you snap a silent photo of a model sticker without speaking, it still reads the label and includes the shot for review. Once you check and approve a quick preview table, SyncWalk saves renamed copies of your shots (like 01_Kitchen_Pipe-Drip.jpg), tags them in UGREEN Files, groups them into a UGREEN Photos album when the folder is indexed, and writes a clean checklist linking every note to its photo and audio timestamp.

How I built it

I built SyncWalk as a Markdown orchestration Skill (sync-walk/SKILL.md with two reference guides) using the official uliya-skill-kit (0.7.7-en), where it passes the official skill checker with zero errors and zero warnings. Instead of bundling custom scripts inside the Skill package, SyncWalk coordinates three built-in UGREEN expert agents: file-agent to list files, read audio length, copy renamed shots in a single batch approval, and apply file tags; knowledge-agent to transcribe speech locally and save the final Markdown checklist; and photo-agent to read photo capture times, inspect images, read text on labels, and create a Photos album.

Challenges I ran into

The first hurdle was file formats. UGREEN's built-in transcription and photo tools read audio and image files directly, so they skip .mp4 and .mov video files. Supporting unpacked video frames plus audio alongside standalone photos plus a .m4a voice memo gives the best of both worlds: single-video walks work cleanly once split by a local UGOS Docker ffmpeg container, while separate photos plus audio work out of the box with zero Docker setup.

Second, matching timestamps across two devices is trickier than it first looks. Many voice recorder apps stamp a file with the time you hit Stop instead of Start, and dedicated cameras often drift by several minutes or use a different timezone setting. Subtracting the audio duration to find the true start time, then double-checking one or two clear photos against what was said, solves both problems reliably.

Third, UGREEN's album tool creates albums from indexed search results instead of raw file paths, and newly copied photos can take a moment to show up in the Photos index. Checking whether the new copies are indexed first lets SyncWalk create the album when ready or skip that step cleanly with a note instead of failing.

Built With

Share this project:

Updates

Submission history