StoryWeaver

Your whole stack. One conversation.

Comment

Inspiration

The Problem That Kept Us Up at Night: As students in tech we've experienced the frustration of joining a new team or inheriting a legacy codebase and spending weeks just trying to understand how everything fits together. Traditional tools give you piecemeal insights - you can see individual files, but you miss the forest for the trees. We wanted to build something that felt like having the world's most experienced senior developer sitting next to you, instantly understanding your entire project and answering any question about it.

What it does

StoryWeaver transforms how developers understand and work with codebases. Unlike traditional AI assistants that analyze single files in isolation, StoryWeaver leverages Gemini's massive 2M token context window to comprehend your entire repository at once.

Core Capabilities:

  • Full Repository Analysis: Instant architecture overviews of 100+ file projects
  • Cross-File Q&A: Ask questions like "How does authentication flow work?" and get answers referencing multiple files
  • Smart Navigation: AI-powered file discovery based on features and concepts
  • Architecture Mapping: Automatic visualization of project structure and dependencies
  • Context-Aware Explanations: Right-click any code for instant, context-rich explanations Real Impact: Developers can understand new codebases in minutes instead of weeks, debug complex issues across multiple files, and make architectural decisions with full context awareness.

How we built it

  • VS Code Extension API for seamless IDE integration
  • Hierarchical context building that prioritizes entry points and key files
  • Language-aware parsers for 15+ programming languages
  • Dependency graph construction to preserve code relationships
  • Structured prompt engineering optimized for code understanding
  • Webview-based UI for rich visualization of analysis results
  • Parallel file processing for performance with large repositories
  • Smart token management to maximize 2M context window usage

Challenges we ran into

  • Token limit management - Even 2M tokens insufficient for massive repos
  • Preserving code structure when flattening 3D codebases into 1D text
  • Performance optimization with 500+ file repositories
  • Natural language understanding for varied developer terminology
  • Error handling for API limits, network issues, and invalid code
  • Memory management during large-scale file analysis
  • User experience design for complex multi-file analysis workflows

Accomplishments that we're proud of

  • Broke the single-file barrier in AI code analysis
  • Created novel UX pattern - "whole codebase conversation"
  • Achieved 95% accuracy on complex multi-file queries
  • Analyzed 150+ file repos in under 15 seconds
  • Built production-ready architecture with error handling
  • Delivered complete MVP within hackathon timeframe
  • Enabled 10x faster onboarding for new developers
  • Reduced cross-file bug resolution time significantly

What we learned

  • Context window ≠ understanding - requires careful prompt engineering
  • VS Code Extension API allows deep but nuanced integration
  • Code has multiple dimensions - structure, dependencies, semantics, evolution
  • Developer tools need instant feedback for user trust
  • Structured prompts outperform large unstructured text dumps
  • Temperature settings matter - low for analysis, higher for suggestions
  • Simple explanations often work better than technical jargon
  • Start with hardest problem - forces core solution development

What's next for StoryWeaver

Short Term (3 Months):

  • VS Code Marketplace release
  • Team collaboration features
  • PR/code review integration
  • Custom architecture rule definitions

Medium Term (6 Months):

  • Multi-repository analysis
  • Visual architecture diagrams
  • Interactive learning mode
  • Voice interface integration

Long Term Vision:

  • Proactive architecture assistant
  • Codebase evolution tracking
  • Cross-language understanding
  • Design system integration

Research Directions:

  • Fine-tuned code understanding models
  • Predictive refactoring suggestions
  • Knowledge graph integration
  • Automated migration planning

Community Goals:

  • Open source core engine
  • Plugin ecosystem development
  • Educational content creation
  • Academic research partnerships
Share this project:

Updates