The Missing Layer: Why AI Memory Should Start with Humans, Not Agents

Everyone is building memory for AI agents. But the richest knowledge — the reasoning traces of humans working with AI — disappears after every session. Doriana will show how she built Luminaria Memo: a live system that captures AI sessions into a Neo4j knowledge graph and reinjects context into new conversations via the Model Context Protocol (MCP).

You will learn:
• The graph data model: how sessions, messages, tags, notes, and links map to nodes and typed relationships
• Hybrid search: semantic search via Neo4j vector indexes (cosine + recency weighting), plus keyword and multi-parameter filtering — vector ranking runs inside Neo4j
• How MCP context injection turns a stateless assistant into one that remembers past decisions, patterns, and mistakes
• Why graph retrieval surfaces what flat vector search misses: multi-hop traversal across sessions, via shared tags, finds connections isolated chunks can't
• Where this is heading: from a personal knowledge graph to shared organizational intelligence — querying a teammate's reasoning traces when they're unavailable, short on time, or simply can't explain it as clearly as their graph can

This isn't a conceptual talk: Doriana will demo live queries on a production Neo4j instance with real data, show the Cypher behind the search pipeline, and walk through the graph model. She'll also share lessons from building a graph-native SaaS solo — what worked, what broke, what she'd do differently.

Attendees will leave with a concrete architecture for AI memory on Neo4j, and a new way to think about who the most important node in the graph really is.