Lost property operators at TFGM manually classify each item found across the network, selecting Subcategory, Category, Colour, Damage, and Brand from dropdowns. With hundreds of items processed daily, manual classification is slow, inconsistent, and error-prone. Operators write natural descriptions but spend minutes hunting through Dataverse dropdowns for the right values.
Let the operator describe the item in their own words, exactly as they found it. AI handles the translation to structured Dataverse data instantly. Staff focus on the passenger, not the dropdown.
— Rahul Bachina
The LP AI Classifier uses a GPT-4.1 AI Hub prompt (LP-Item-Classifier) triggered by an operator typing a free-text item description in Power Pages. A Power Automate HTTP flow passes the description plus JSON arrays of valid Dataverse lookup values to the AI prompt. The model performs fuzzy/similarity matching, tolerant of typos, synonyms, and informal wording, and returns a structured JSON response that auto-fills the Lost Property form fields, reducing manual classification time and improving data consistency.
- GPT-4.1 AI Hub prompt with fuzzy matching tolerant of typos, synonyms, and informal wording
- Auto-fills five fields simultaneously from a single free-text description
- Integrated directly into Power Pages via a Power Automate HTTP flow
- Dataverse lookup resolution returns accurate subcategory and category record IDs
- Reduces item classification from minutes to under three seconds
Want to collaborate or learn more?
Reach out to discuss this project, a partnership, or a new idea.
Get in Touch