Code Intelligence Graph

Overview

Every merge request is reviewed against a graph of the whole codebase, not just the diff. The graph knows which functions call which, where data flows, and which modules are owned by whom.

The problem

Diff-only review misses the call that changed three files away. Reviewers were spending their time rebuilding context that a machine already had.

What I built

• Architected the ingestion pipeline on Temporal, with incremental re-indexing on every push so the graph stays current.
• Designed the graph schema around call edges, data flow edges, and ownership, with per tenant isolation enforced at the database layer.
• Implemented the review agent that walks the graph from the diff outward and writes findings with file and line references.
• Resolved Temporal heartbeat timeouts caused by unbounded Neo4j dependency queries on very large graphs, stabilizing the ingestion pipeline.
• Drove security hardening across the platform, covering encrypted connections and encryption at rest.

Impact

• Reviewers get findings with file and line references instead of rebuilding context by hand.
• Security-focused reviews run with tenant isolation and encryption, so customer code stays separated and protected.