Defect Solver 🪲

AI-powered bug localization for microservice architectures using hierarchical code understanding and LLM-based analysis.

⚠️ Disclaimer: Some links point to private repositories accessible only to authorized team members.

PROJECT INDEX:

1. Defect Solver API: locates buggy repos and files
2. Defect Solver Codebase Summarizer: converts codebases to NL knowledge base
3. Dnext Coder MCP: a gateway to acccess our API
4. Central Storage: stores NL knowledge bases
5. Defect Solver Agent: a simple code agent to fix bugs

What It Does

Provide a bug description in natural language, get back a ranked list of microservices and files likely containing the defect.

Example:

User: "Why is the user profile not loading?"
Defect Solver: Bug likely in user-service microservice, file UserProfileController.java

How It Works

The system uses MCP (Model Context Protocol) to integrate with AI development environments:

flowchart LR
    User -->|Bug Description| Agent
    Agent -->|MCP Tool| Server[MCP Server]
    Server -->|API Call| DS[Defect Solver API]
    DS -->|Ranked Results| Agent

Loading

See the High Level Diagram for a detailed system overview.

Quick Start

1. Connect to MCP Server - Use hosted version at https://dnext-coder-mcp-server.pia-team.com/mcp/
2. Configure Your IDE - VSCode, JetBrains, Claude Desktop, or any MCP-compatible environment
3. Follow Setup Guide - See User Guide for configuration and usage
4. Configure AI Agent - Copy AGENTS.md to your project for optimal agent behavior

Architecture

Two-phase pipeline: Search Space Routing → Bug Localization

1. Phase 1: Identify top-N suspicious microservices from bug description
2. Phase 2: Pinpoint top-M suspicious files within selected microservices

See Algorithm Details for complete breakdown.

Development & Deployment

• For contributors and developers, see Developer Guide.
• For deployment instructions, see Deployment Guide.

Research

Based on hierarchical code understanding research papers to overcome LLM context window limitations in large-scale projects.

Paper 1: Repository-Level Code Understanding by LLMs via Hierarchical Summarization

Paper 2: Natural Language Summarization Enables Multi-Repository Bug Localization by LLMs in Microservice Architectures