Building a Graph-Based Pattern Detection System: What I Learned and Where It Led

I built Ascent Ledger as a career diagnostic OS — graph-based pattern detection on professional trajectories. The product taught me more about AI system architecture than almost anything else I built. This is the technical story — what the graph approach unlocked, what it cost, and how the thinking transferred directly into PRISM and NexOps. Why Graph Over Vector for Pattern Detection The limitation of vector similarity for career data: Vectors find similarity, graphs find structure A career trajectory is not a set of similar documents. It is a sequence of connected decisions with causal relationships Why FalkorDB: native graph queries, relationship traversal, pattern matching across nodes Code snippet: basic graph schema for career nodes and edges The Pattern Recognition Layer What a "stall pattern" looks like in graph form vs in a CV How the system detects structural loops — the same role type, different company, no progression The difference between movement and ascent — the insight