--- title: AI on Your Lakehouse: Context Comes in Shapes, Not... description: A workshop transcript covering how to use Neo4j graph databases to improve AI agent context retrieval in a lakehouse architecture. The session introduces three... canonical: https://daily.dev/posts/ai-on-your-lakehouse-context-comes-in-shapes-not-queries-zach-blumenfeld-neo4j-wsecwsrwl twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: AI on Your Lakehouse: Context Comes in Shapes, Not Queries — Zach Blumenfeld, Neo4j | daily.dev og:description: A workshop transcript covering how to use Neo4j graph databases to improve AI agent context retrieval in a lakehouse architecture. The session introduces three... og:url: https://daily.dev/posts/ai-on-your-lakehouse-context-comes-in-shapes-not-queries-zach-blumenfeld-neo4j-wsecwsrwl og:image: https://api.daily.dev/og/posts/WsecwsrWl.png og:image:alt: AI on Your Lakehouse: Context Comes in Shapes, Not Queries — Zach Blumenfeld, Neo4j og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # AI on Your Lakehouse: Context Comes in Shapes, Not Queries — Zach Blumenfeld, Neo4j **[AI Engineer](https://daily.dev/sources/aidotengineer)** · 119 min read · 0 upvotes · 0 comments ## Summary A workshop transcript covering how to use Neo4j graph databases to improve AI agent context retrieval in a lakehouse architecture. The session introduces three 'context shapes': a connection/semantic layer on top of BigQuery structured data (using NeoArta to build a metadata graph), a table-of-contents/outline shape for navigating unstructured documents (PDFs, manuals, bulletins), and a themes/communities shape for surfacing global patterns. The workshop uses Claude Code as the AI agent, Cypher as the graph query language, and demonstrates how graph representations help agents answer estate-level questions, prove negatives, and traverse large document stores more reliably than vector search alone. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How can an AI agent answer estate-level questions like 'what documentation are we missing' when vector search can't prove a negative? Vector and lexical search can only match things that exist, so they cannot confirm absence. A graph-based approach uses community detection (the Leiden algorithm) to cluster interlinked documents into themes purely from document structure and link metadata, then correlates those themes against warehouse work-order history to surface DTC codes or repair patterns with no matching documentation. _For teams weighing graph versus vector approaches for agent context, daily.dev surfaces the tradeoffs as they emerge._ ### What is the difference between an ontology and a semantic layer when giving an AI agent access to a data warehouse? An ontology helps a system interpret and reason about data, while a semantic layer helps it understand the consistent, agreed-upon terms needed to query data accurately. A tool like Neo4j's Neocarta builds a metadata graph over warehouse tables (using primary/foreign key joins from the information schema) so an agent can correctly join hundreds of tables without copying the underlying data via ETL. _Developers designing agent-ready data layers can track evolving semantic-layer patterns on daily.dev._ ### Why might a deterministic document-loading approach be preferred over LLM-based entity extraction when building a graph for RAG? A deterministic load builds the graph purely from existing document structure (folders, sections, and named links) rather than using an LLM to extract entities, making it idempotent, faster, and cheaper to run repeatedly. The tradeoff is it depends on documents already being well-structured and consistently named, since messier sources may still require regex, NLP, or LLM-assisted parsing. _Anyone evaluating lightweight versus entity-extraction RAG pipelines can follow related approaches on daily.dev._ ## Similar posts on daily.dev - [Why AI agents need an AI lakehouse in the modern enterprise](https://daily.dev/posts/why-ai-agents-need-an-ai-lakehouse-in-the-modern-enterprise-xux8icgxw) · Dynatrace · 3 upvotes · 0 comments - [“A data lake of nuance for AI agents to swim in”: AWS Context gets shipshape on reasoning](https://daily.dev/posts/a-data-lake-of-nuance-for-ai-agents-to-swim-in-aws-context-gets-shipshape-on-reasoning-b2kax0psv) · The New Stack · 0 upvotes · 0 comments - [Governance beyond security: knowledge, context & ontology on the lakehouse](https://daily.dev/posts/governance-beyond-security-knowledge-context-ontology-on-the-lakehouse-ohjifsyku) · databricks · 1 upvotes · 0 comments - [Graph Databases \+ LLMs: Why Connected Data Makes AI Smarter • Arango](https://daily.dev/posts/graph-databases-llms-why-connected-data-makes-ai-smarter-arango-gosgr4cvj) · ArangoDB · 1 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#architecture](https://daily.dev/tags/architecture), [#data-engineering](https://daily.dev/tags/data-engineering), [#neo4j](https://daily.dev/tags/neo4j) [View this post on daily.dev](https://daily.dev/posts/ai-on-your-lakehouse-context-comes-in-shapes-not-queries-zach-blumenfeld-neo4j-wsecwsrwl) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"AI on Your Lakehouse: Context Comes in Shapes, Not Queries — Zach Blumenfeld, Neo4j","url":"https://daily.dev/posts/ai-on-your-lakehouse-context-comes-in-shapes-not-queries-zach-blumenfeld-neo4j-wsecwsrwl","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/ai-on-your-lakehouse-context-comes-in-shapes-not-queries-zach-blumenfeld-neo4j-wsecwsrwl"},"datePublished":"2026-07-23T07:23:24.941Z","dateModified":"2026-09-14T08:03:19.927Z","description":"A workshop transcript covering how to use Neo4j graph databases to improve AI agent context retrieval in a lakehouse architecture. The session introduces three...","image":"https://i.ytimg.com/vi/kRkcNOsRyYg/sddefault.jpg","thumbnailUrl":"https://i.ytimg.com/vi/kRkcNOsRyYg/sddefault.jpg","isAccessibleForFree":true,"articleSection":"AI Engineer","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"AI Engineer","logo":"https://media.daily.dev/image/upload/s--u5PucxNT--/f_auto/v1724338940/logos/aidotengineer","url":"https://daily.dev/sources/aidotengineer"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/ai-on-your-lakehouse-context-comes-in-shapes-not-queries-zach-blumenfeld-neo4j-wsecwsrwl","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,architecture,data-engineering,neo4j","timeRequired":"PT119M","video":{"@type":"VideoObject","name":"AI on Your Lakehouse: Context Comes in Shapes, Not Queries — Zach Blumenfeld, Neo4j","description":"A workshop transcript covering how to use Neo4j graph databases to improve AI agent context retrieval in a lakehouse architecture. The session introduces three...","thumbnailUrl":"https://i.ytimg.com/vi/kRkcNOsRyYg/sddefault.jpg","uploadDate":"2026-07-23T07:23:24.941Z","duration":"PT119M","url":"https://api.daily.dev/r/WsecwsrWl","embedUrl":"https://www.youtube.com/embed/kRkcNOsRyYg"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"AI Engineer","item":"https://daily.dev/sources/aidotengineer"},{"@type":"ListItem","position":3,"name":"AI on Your Lakehouse: Context Comes in Shapes, Not Queries — Zach Blumenfeld, Neo4j"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/ai-on-your-lakehouse-context-comes-in-shapes-not-queries-zach-blumenfeld-neo4j-wsecwsrwl#faq","mainEntity":[{"@type":"Question","name":"How can an AI agent answer estate-level questions like 'what documentation are we missing' when vector search can't prove a negative?","acceptedAnswer":{"@type":"Answer","text":"Vector and lexical search can only match things that exist, so they cannot confirm absence. A graph-based approach uses community detection (the Leiden algorithm) to cluster interlinked documents into themes purely from document structure and link metadata, then correlates those themes against warehouse work-order history to surface DTC codes or repair patterns with no matching documentation. For teams weighing graph versus vector approaches for agent context, daily.dev surfaces the tradeoffs as they emerge."}},{"@type":"Question","name":"What is the difference between an ontology and a semantic layer when giving an AI agent access to a data warehouse?","acceptedAnswer":{"@type":"Answer","text":"An ontology helps a system interpret and reason about data, while a semantic layer helps it understand the consistent, agreed-upon terms needed to query data accurately. A tool like Neo4j's Neocarta builds a metadata graph over warehouse tables (using primary/foreign key joins from the information schema) so an agent can correctly join hundreds of tables without copying the underlying data via ETL. Developers designing agent-ready data layers can track evolving semantic-layer patterns on daily.dev."}},{"@type":"Question","name":"Why might a deterministic document-loading approach be preferred over LLM-based entity extraction when building a graph for RAG?","acceptedAnswer":{"@type":"Answer","text":"A deterministic load builds the graph purely from existing document structure (folders, sections, and named links) rather than using an LLM to extract entities, making it idempotent, faster, and cheaper to run repeatedly. The tradeoff is it depends on documents already being well-structured and consistently named, since messier sources may still require regex, NLP, or LLM-assisted parsing. Anyone evaluating lightweight versus entity-extraction RAG pipelines can follow related approaches on daily.dev."}}]} ```