Beyond Vibe Coding: AI Agents for Technical Debt Detection with Neo4j

Session Track: App Dev

Session Time: 11:30 pm - 12:00 am November 7

Session description

In the fast-paced world of software development, "vibe coding"—where intuition and momentum drive decisions—can lead to rapid progress but often accrues unseen technical debt. This session unveils a suite of AI agents I've developed to detect and analyze the subtle signs of such debt using Neo4j's graph capabilities. Code Intelligence Architecture with Neo4j I've built an extensible framework using Dagger and Neo4j that: Parses diverse codebases (Python, JavaScript, TypeScript, etc.) using specialized parsers Constructs a comprehensive code graph in Neo4j including: File nodes with language and path metadata Symbol nodes (Functions, Classes, Interfaces) with scope, signature, and line positioning Relationship edges: DEFINED_IN, IMPORTS, CALLS, CONTAINS, and re-export patterns Executes Cypher queries through a dedicated cypher-shell client container Live Demonstration During this talk, I'll demonstrate: Building a code graph for a real-world repository: # Creating file and symbol nodes with proper metadataawait neo_service.add_file_node(filepath="/app/src/hooks/use-toast.ts", language="typescript")await neo_service.add_symbol(    symbol_type="Function",    name="useToast",    filepath="/app/src/hooks/use-toast.ts",    start_line=10,    end_line=45,    properties=SymbolProperties(scope="public")) Detecting code smells through graph pattern analysis: # DeadCodeDetector query demonstrating pattern detection for unused exportsMATCH (symbol:Function|Class|Interface)-[:DEFINED_IN]->(f:File)WHERE (symbol.scope IS NULL OR symbol.scope = "public")AND NOT EXISTS {    # Check if the file is imported directly    MATCH (other:File)-[:IMPORTS]->(f)    WHERE other  f}# Additional pattern matching for re-exports and references Concurrent multi-detector analysis with error handling: Circular dependencies Dead code Shotgun surgery Feature envy Schema discrepancies Takeaways Attendees will learn how to: Model code structures effectively as property graphs in Neo4j Write advanced Cypher queries to detect complex code patterns Integrate Neo4j with AI agents for automated code analysis Handle property existence checks and schema evolution challenges Build resilient code analysis tools that can adapt to different codebases Whether you're a developer, team lead, or architect, this talk offers practical strategies to harness Neo4j and AI for proactive code health monitoring, ensuring the creative spark of vibe coding doesn't compromise long-term maintainability.