How I'm using ASTs and Gemini to solve the "Codebase Onboarding" problem đź§
gemini
| Source: Dev.to | Original article
Tara Mäkinen, a senior software engineer and consultant, has unveiled a practical workflow that blends abstract syntax trees (ASTs) with Google’s Gemini model to cut the learning curve for developers joining large codebases. In a detailed post published today, she explains how her consultancy tool, AuraCode, automatically extracts ASTs from a repository and feeds them into Gemini’s long‑context window, letting the model generate a structured onboarding guide in minutes rather than days.
For small‑to‑medium projects, AuraCode injects the full AST directly into Gemini’s context, enabling the model to answer granular questions about function signatures, data flows and architectural patterns. In larger monorepos, the tool first partitions the AST into thematic chunks—e.g., UI layer, data access, build scripts—and uses Gemini’s summarisation capabilities to stitch together a high‑level overview before drilling down on demand. The result is a two‑tier guide that combines a concise architecture map with line‑by‑line explanations, all kept up‑to‑date as the code evolves.
As we reported on 15 April, Tara’s initial experiments demonstrated that Gemini could turn raw code into readable documentation, but the new post adds the scaling logic that makes the approach viable for enterprise‑size repositories. The method sidesteps the chronic problem of stale READMEs and scattered Confluence pages, offering a dynamic, AI‑driven alternative that can be regenerated with each commit.
The significance extends beyond onboarding. Continuous generation of AST‑enhanced prompts could feed into automated code reviews, security audits and even test‑case synthesis, turning Gemini into a multi‑purpose assistant for the entire development lifecycle.
Watch for the upcoming open‑source release of AuraCode’s AST extraction pipeline, slated for early May, and for Google’s next Gemini update, which promises an even larger context window and native AST awareness. Together they could set a new standard for AI‑augmented software engineering in the Nordics and beyond.
Sources
Back to AIPULSEN