I'm Shivansh, and I built CodeContext to solve a problem every growing engineering team faces: new developers taking 2-3 months before their first meaningful contribution.
The core insight: Most onboarding time isn't spent learning to code—it's spent figuring out which 15 files (out of 10,000) actually matter, and in what order to read them.
What CodeContext does:
Scans your repository (any language)
Builds a dependency graph using AST parsing
Runs PageRank to identify "knowledge hotspots" (the UserService.java that's called by 47 other files)
Generates a topologically-sorted "reading list" (simple → complex)
Creates interactive D3.js visualizations
Example output:
text Your Codebase Map ━━━━━━━━━━━━━━━━━━━━━━━ Knowledge Hotspots: 1. AuthService.kt (89 connections) - CRITICAL 2. DatabaseConfig.kt (67 connections) 3. UserRepository.kt (45 connections)
Week 1 Onboarding Path: → README.md, ARCHITECTURE.md → Models: User.kt, Product.kt (low complexity) → Config: application.yml, DatabaseConfig.kt → First task: Add validation to UserDTO Technical stack:
Kotlin (multiplatform-ready)
Tree-sitter for parsing 40+ languages
JGraphT for graph algorithms
D3.js for interactive visualizations
Git history analysis for change patterns
Why this approach works:
Traditional docs go stale immediately
Code search is great if you know what to search for
AI assistants generate code but don't teach architecture
Juniors need a "start here" button, not a search box
What's different from Sourcegraph/CodeSee:
Lightweight (single CLI command, no infrastructure)
Onboarding-first (not code search)
Free & open source (MIT license)
Runs locally (no code leaves your machine)
I've been testing this with 3 teams (50-200 developers). Average time to first PR went from 47 days → 19 days.
Try it:
bash npm install -g codecontext codecontext analyze /path/to/repo Open to feedback! What features would make this immediately useful for your team?