Using BigQuery Graph with Kineviz GraphXR

Over 80% of enterprise data lives in unstructured form — PDFs, emails, reports, regulatory filings. Most of the time, such sources contain critical business information, yet they remain difficult to access and reason over at scale. Together, BigQuery Graph and Kineviz GraphXR give decision makers power over their unstructured data by creating a single, streamlined workflow that makes it much easier to uncover hidden business insights. BigQuery houses and builds the structures of the graph; Kineviz GraphXR lets analysts visually verify relationships, trace insights back to sources, and answer questions interactively. 

Retrieval-augmented generation (RAG) and vector search have become the industry standard approach for working with unstructured data. When it comes to trend analysis, comparison across entities, multi-hop reasoning, and explainable decision support, graphs complement RAG by incorporating context and relationship mapping. Our “evidence-first” knowledge graph approach prioritizes preserving the nuance of the original evidence and maintaining the traceability of every single element in the graph, making the resulting analysis verifiable and trustworthy. In this post, we describe an example where BigQuery AI Functions, BigQuery Graph, and Kineviz GraphXR address business questions about Fortune 500 SEC filings without complex ETL pipeline, data duplication, or separate graph databases. 

From fragmented to unified with  BigQuery

Traditional unstructured analytics pipelines can be complex and sprawling. They typically involve multiple steps, including: object storage for raw files, a custom parsing service, a separate AI extraction layer, a standalone graph database, and finally, a BI tool for analysis. This complex setup can be difficult to maintain, involving data duplication, synchronization overhead, introducing multiple potential points of failure.

BigQuery streamlines this process. Raw documents are stored in Google Cloud Storage, and text extraction, Gemini-powered inference, and graph creation all run directly within the same platform. This removes the need for data movement between systems, complex service orchestration, or the accumulation of out-of-sync data copies.

With its tight integration, the pipeline is simple and maintainable, allowing full provenance without bespoke infrastructure.