The editorial team of Nordlicht Reisen GmbH (R-01) wants to find hotel photos and short hotel clips (R-04) in the travel assistant's search: "Show hotels with a pool" should hit photos even if the word is missing in the file name, and for a clip the portal should jump to the place where the pool is visible. Content Understanding turns images and videos into text and fields that an index accepts (lesson 5.1). Paper: requirements, decision, field definition, compiled code; only the optional cloud step needs an Azure account. Cost drivers: per image, per minute of video and the tokens of the model behind it (billed on your deployment according to the pricing page).
Prior knowledge. Ingesting content and preparing it for search is in "AI track, lesson T4.3", choosing services in "lesson 1.1" and Content Understanding as an agent's tool in "lesson 2.4" of this course. For alt texts and questions about a single photo "lesson 3.3" applies; here it is about cataloging at scale.
The requirement
All numbers are assumptions of the task.
| No. | Requirement | Type | Measure | Weight | Source |
|---|---|---|---|---|---|
| 3.4-01 | Every hotel photo gets a description for the index. | F | 1 paragraph per photo, 100 % of photos | Must | Editorial team |
| 3.4-02 | Hotel clips are split into segments with timestamps and description. | F | segment with start and end | Must | Editorial team |
| 3.4-03 | Custom fields per photo: subject, visible things, interior yes or no. | F | 3 fields | Should | Editorial team |
| 3.4-04 | Jobs run only if file, length and dimensions fit the limits. | R | 0 jobs rejected by the service | Must | Operations |
| 3.4-05 | The location of processing is the EU (R-06). | R | region per resource | Must | Data protection |
| 3.4-06 | Objects with their position in the image (bounding box) for a marking in the portal. | F | bounding box per object | Could | Editorial team |