Distributed multimedia knowledge-graph pipeline for entity and relationship extraction
Process images, videos, audio, and text from Cloud Storage with distributed producer and worker jobs, use Gemini to extract structured entities and relationships, store the results in BigQuery tables, and expose them as a queryable property graph.
From Google Cloud Tech — How to build and scale multi-agent AI systems on GKE at 01:03:42
Problem: Semantic search and embeddings can retrieve similar information but are limited when users need to traverse relationships between entities, determine impact across connected systems, or answer multi-hop questions. Processing large multimedia collections also creates a compute and throughput challenge.
For: Organizations with large multimedia collections or complex domain data—such as manufacturing companies managing manuals, policies, parts, quality assessments, and customer feedback—that need relationship-aware search and analysis.
Examples
- The pet-media pipeline extracted facts from pictures, videos, audio, and text about pets' breeds, hobbies, favorite foods, and preferred toys.