Using ASTs and Gemini for Codebase Onboarding with AuraCode

说明与用途

Tara 开发了 AuraCode，这是一个使用 AST 和 Gemini 创建交互式代码架构图的开发人员入职工具。该工具通过基于代码结构生成可视化和上下文感知响应，解决了理解复杂系统的挑战。

问题

New engineers take 3 to 6 months to understand complex systems, spending nearly 50% of their time reading and navigating code. Existing AI tools focus on writing code but fail to solve the comprehension layer for undocumented or legacy repositories.

结果

AuraCode transforms undocumented repositories into interactive visual maps and provides context-aware chat. It uses ASTs for smaller repos and Lean RAG for larger ones to maintain accuracy while managing token costs. The tool generates onboarding checklists and PR summaries, aiming to reduce the 3-6 month onboarding period.

工作流

• Paste a GitHub URL into AuraCode.
• For small/medium repos: Inject Abstract Syntax Trees (ASTs) into the model's context window to preserve architectural patterns.
• For massive monorepos: Use 'Lean RAG' to selectively surface structurally relevant nodes before context injection to manage token limits.
• Generate an interactive D3.js radial tree mapping component relationships.
• Query the system via chat for context-aware answers (e.g., impact analysis of code changes).
• Optionally use ElevenLabs TTS to hear the flow explained hands-free.
• Review auto-generated onboarding checklists tailored to the scanned architecture.

证据

it takes new engineers 3 to 6 months to truly understand how a complex system is glued together

来源

AuraCode uses two different approaches depending on the size of the repository: For small/medium repos: I inject Abstract Syntax Trees (ASTs)... For massive monorepos: I use what I call Lean RAG.

来源