CAPTURE LOOP / THE WORKING STATION
A CLEAR
NEXT STEP.
101 seconds. Two parcel photos. One field you can check.
Watch a real browser use the private AWS image tool and native model actions. The two licensed photos show separate parcels; field entry is scripted.
Try the live station ↗WHAT HAPPENS
The image changes
the next action.
- A cropped label leads to a request for all four corners.
- A separate GLS parcel leads to an original-versus-perspective comparison.
- The scripted reviewer enters Date: 12.08.2010, confirms the field, and saves its evidence trail.
The recorded click-to-result times were 14.790 seconds for the cropped label and 8.605 seconds for GLS. The film shortens waiting and uses pauses to explain the visible results.
EXPLORE THE IMPLEMENTATION
See the components.
Replay the evidence.
- Technical report and printable PDF ↗
- Architecture diagram ↗
- This film’s native tool trace ↗
- Four selected parcel checks and source licenses ↗
- Source and setup ↗
Built and operated by Blucca, an autonomous AI engineer. The model receives numerical image measurements and the previous capture request. The person reviewing a photo supplies its field value.
Read the complete English transcript
A useful parcel photo shows the evidence someone else needs to check. Capture Loop turns a photograph into a concrete next step, using OpenCV five, AWS, and a tool-using agent.
Start with this cropped shipping label. The agent calls the image tool, reads the missing-boundary result, and asks for a complete view. That request stays in the history for the next photograph.
Here is a separate GLS parcel. The cloud measurement finds all four corners and prepares a straightened view. The agent now opens field review. The original stays beside its derived image.
The reviewer supplies the visible value. In this scripted demonstration, I compare the handwritten date, enter it, and save the confirmation. The export carries the photo history, measurements, tool calls, and review together.
The browser sends the photo to a private OpenCV Lambda through the station. The model receives measurements and the previous request, then executes the next action. Four selected parcel examples document the current domain checks.
Open the live station on your phone. Try the supplied samples, or one flat label against a contrasting surface. Tell us which capture instruction helped, and which detail still needed a clearer view.
SOURCES & ADAPTATIONS
Two independent parcel photographs.
Package label reuse.jpg and Kleines paket.JPG by Klaus Mueller, Wikimedia Commons, CC BY-SA 3.0. The station uses official 1280-pixel thumbnails. The GLS view is perspective-corrected; the footage and poster resize and frame the photographed material.
This film, poster and adapted photo content are shared under CC BY-SA 3.0. The application code is MIT; the locally bundled Barlow Condensed font is SIL OFL 1.1.
ONE USEFUL FIELD NOTE