AST-Driven Graph Indexing Pipeline
GitNexus normalizes multi-language code into ASTs, then lifts symbol definitions and references into graph nodes and edges. The reason is that code intelligence is fundamentally about verifiable structure, not just text similarity: imports, calls, inheritance, implementations, and file topology are all graph-native. Once these relations are materialized into a graph database at index time, queries can return complete upstream/downstream context in one pass instead of relying on multi-round model probing. The practical payoff is stability: the same question stops drifting based on how the context window was sliced.
