# Building a Multimodal LLM Pipeline for Structured Visual-Asset Metadata

DevFeed: [Building a Multimodal LLM Pipeline for Structured Visual-Asset Metadata](<https://devfeed.tech/articles/teaching-a-vision-model-to-see-like-a-human-annotator-and-catching-it-when-it-lies-50201.md>)

Original publisher: [Read original article](<https://devblogs.microsoft.com/ise/ai-asset-enrichment-pipeline/>)

Author: Deeptanil Saha, Kartheek Palepu

Published: 2026-07-17T07:00:00Z

Content type: article

Language: en

Sources: [ISE Developer Blog](<https://devfeed.tech/sources/ise-developer-blog.md>)

Topics: [multimodal-ai](<https://devfeed.tech/topics/multimodal-ai.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Ground truth / benchmark quality](<https://devfeed.tech/topics/ground-truth-benchmark-quality.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>)

Tags: [cognitive-services](<https://devfeed.tech/tags/cognitive-services.md>), [cse](<https://devfeed.tech/tags/cse.md>), [developer](<https://devfeed.tech/tags/developer.md>), [embeddings](<https://devfeed.tech/tags/embeddings.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [hallucinations](<https://devfeed.tech/tags/hallucinations.md>), [ise](<https://devfeed.tech/tags/ise.md>), [llm](<https://devfeed.tech/tags/llm.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [model](<https://devfeed.tech/tags/model.md>), [multimodal](<https://devfeed.tech/tags/multimodal.md>), [pipeline](<https://devfeed.tech/tags/pipeline.md>), [structured](<https://devfeed.tech/tags/structured.md>), [vision](<https://devfeed.tech/tags/vision.md>)

## AI overview

This article describes a multimodal LLM pipeline that extracts structured metadata from visual assets for production search. It uses constrained values, image-first grounding, preprocessing, and evaluation to reduce hallucinations and improve trust in the resulting metadata.

## Source excerpt

A multimodal LLM enrichment pipeline that extracts structured metadata from visual assets, constrains output to predefined values to minimize hallucinations, and uses a ground truth evaluation template to measure quality--all as a plug-and-play module. The post Teaching a Vision Model to See Like a Human Annotator--and Catching It When It Lies appeared first on ISE Developer Blog.