This Week in Neo4j: Agent Orchestration, GraphRAG, Geospatial Knowledge Graph and more

FEATURED COMMUNITY MEMBER: Pratiksha Zalte

AGENT ORCHESTRATION: Using Neo4j Change Data Capture to Orchestrate Agents

Instead of adding a message broker next to your database, Julian Busch lets the graph write it itself: Change Data Capture turns a task moving from PENDING to ACTIVE into the signal that wakes the next agent. State, handoffs and the audit trail stay connected in a single graph and even failed events become nodes you can triage with Cypher. He is open about the trade-offs (roughly 750 ms detection delay per hop, a limited replay window) and shows when it’s time to move on to Kafka.

GRAPHRAG: Hybrid GraphRAG framework for semantic integration and querying of levee infrastructure data

Armita Davarpanah and colleagues built a levee knowledge graph in Neo4j from public U.S. Army Corps of Engineers data, mapped to a formal ontology and made it queryable in plain language. On their own benchmark of entity, attribute, relationship and counting questions, they report 95.16% exact-match accuracy, with each answer traceable to source records. It’s a useful example of ontology-first graph modeling in a safety-critical domain.

GEOSPATIAL: Real-Time Vehicle Tracking With Neo4j, Databricks Lakebase, and OpenStreetMap

Akmal Chaudhri splits a fleet dashboard across different databases. Neo4j Aura holds the road network as a graph of intersections and roads and answers shortest-path and nearest-intersection questions in Cypher. Databricks Lakebase takes the live vehicle positions and  Lakehouse handles history. The most interesting part is a cross-system join that finds the busiest named roads by combining position data with road names from the graph. A single YAML file switches the whole setup from London to San Francisco or Singapore.

KNOWLEDGE GRAPH: Cogito Estella

Jeffrey Romero built Cogito Estella, which is an open-source engine that extracts a knowledge graph from text without an LLM – one non-autoregressive forward pass per sentence on consumer hardware. It ships a 3-line connector that MERGEs triples into Neo4j with per-edge provenance, idempotent re-ingestion and cross-lingual node fusion.

STARTUPS: Botman AI

Alejandro D’Andrea spent his career as a CIO fighting legacy systems, including a stint running IT for Walmart Chile. Now he’s the founder and CTO of BotMan-AI, which autonomously migrates entire legacy codebases (Delphi, Visual Basic, Oracle Forms) to modern architectures. A graph of every module, routine, table, and dependency hands each agent exactly the context it needs. One customer’s UI forms each touch 700 files; the graph tracks them all. D’Andrea says the platform has migrated tens of legacy systems so far. Watch the interview

Are you building on graphs? The Neo4j Startup Program gives Series B and earlier startups up to $16K in Aura credits, time with graph engineers, and co-marketing, such as the interviews above. Apply now

POST OF THE WEEK: Ammaar N.

Please share it if you like it!