Skip to content

Backend Overview

The DevFlow backend is a Python-based service that powers the AI-driven code search and understanding features. It handles code parsing, indexing, embeddings, and provides RESTful APIs for the VS Code extension.

What the Backend Does

  • Code Parsing: Analyzes source code files and extracts functions, classes, and documentation.
  • Vector Embeddings: Generates semantic embeddings for code snippets using advanced language models.
  • Search & Retrieval: Provides fast, semantic search across your entire codebase.
  • AI Integration: Connects with OpenAI APIs for context-aware code explanations and answers.
  • Data Management: Stores code metadata and embeddings in a vector database (ChromaDB).

Architecture

The backend follows a modular architecture with clear separation of concerns:

backend/
├── app/
│   ├── api/           # REST API endpoints
│   ├── core/          # Configuration and utilities
│   ├── db/            # Database models and connections
│   └── services/      # Business logic services
├── data/              # Data storage
├── tests/             # Test suite
└── requirements.txt   # Python dependencies

Key Components

API Layer (app/api/)

  • endpoints.py: Main REST API endpoints for search, indexing, and health checks.
  • Handles HTTP requests and responses.
  • Provides OpenAPI/Swagger documentation.

Core (app/core/)

  • config.py: Application configuration and environment variables.
  • utils.py: Utility functions and helpers.

Database (app/db/)

  • metadata_store.py: Manages file metadata and indexing information.
  • vector_store.py: Handles ChromaDB operations for embeddings.
  • chroma_db/: Vector database for storing code embeddings.

Services (app/services/)

  • code_parser.py: Parses and analyzes source code files.
  • embedder.py: Generates vector embeddings for code snippets.
  • rag_engine.py: Retrieval-Augmented Generation for AI answers.
  • cache.py: Caching layer for improved performance.
  • rate_limiter.py: API rate limiting and throttling.

Data Flow

  1. Indexing: Code files are parsed, chunked, and embedded into vectors.
  2. Search: User queries are embedded and matched against stored vectors.
  3. Retrieval: Relevant code snippets are retrieved and ranked.
  4. AI Processing: Context is sent to OpenAI for enhanced answers.

Technology Stack

  • FastAPI: Modern Python web framework for APIs.
  • ChromaDB: Vector database for embeddings storage.
  • OpenAI: AI language models for code understanding.
  • Pydantic: Data validation and serialization.
  • Uvicorn: ASGI server for production deployment.

The backend is designed to be lightweight, fast, and easily extensible for custom use cases.