Skip to content
cd ../projects

ai // project

Skannr

Universal AI Code Analyzer

skannr-ten.vercel.app source
Screenshot of Skannr

## what I built

  1. 01

    Built a TypeScript-based universal code analysis engine that scans large codebases and generates compressed structural "skeletons," cutting token usage by ~96.5% versus full-file scans for AI agent context.

  2. 02

    Designed a hybrid file-ranking system combining lexical matching, import/export graph structure, and dependency-graph centrality with a cross-rerank pass, achieving ~75% top-1 keyword coverage on real-world queries.

  3. 03

    Shipped both a CLI and an MCP (Model Context Protocol) server integration for Gemini CLI, exposing a scan_codebase tool with grounded, evidence-cited responses and a hybrid in-memory/disk cache layer.