Soulbound Codex

landing page

SoulBound Codex

Where ancient powers converge with modern intelligence. Connect with historical minds, orchestrate multi-agent workflow. The codex remembers all who enter.

Inspiration

Our project came from wanting to make productivity tools more engaging with combining various technlogies and features to create something powerful. We asked: what if for this challenge, we built something we would use and practical to our daily life. That's when we decided to submit for the Frankentstein challenge and add unique features that come together to make something powerful - an all-in one tool with many features we would use daily!

We drew from:

Gothic aesthetics: dark themes, mystical elements, atmospheric design
Classic productivity apps: Notion, Todoist
Halloween culture: spooky themes, supernatural elements
Game based design: visual feedback, particle effects, physics-based interactions

What it does

🔮 Terminal Tarot Git repo analysis:
- Analyzes 30 days of commit history of a given project GitHub repo
- Generates three-card tarot spreadsand parses project information
- Assessment of coding habit and standards, giving a score for each category
- Personalized score in a weight scoring framwork, based on coding patterns
⚰️ Graveyard Dashboard Task management where tasks appear as tombstones:
- Physics-based animations (wind sway, moon phase lighting)
- Interactive "digging" to reveal task details
- Particle effects (life energy, spirits, urgency indicators)
- Moon phase calendar with astronomical accuracy
- Pomodoro timer integrated with task tracking
✍️ Ghost Writer AI-powered writing assistant:
- Context-aware suggestions powered bymultiple agents
- Cursed typewriter effect with letter-by-letter animation
- Random cursed character transformations
📖 Necronomicon NotesAncient book-styled note-taking:
- Markdown support with live preview
- Full-text search with highlighted results
- Tag-based organization
- Split-view editor
- Cursed typewriter effects

Workflow orchestration for Ghost Writer and Terminal: Fragment Restoration: Multi-agent collaborative restoration of corrupted historical documents Deep Analysis: Chain-of-thought analysis with multiple agents providing perspectives Creative Collaboration: Parallel creative generation, then synthesis Fact Checking: Multiple agents verify information, orchestrator resolves conflicts Codebase Review: Multi-agent comprehensive codebase review with different perspectives (architecture, security, code quality)

Additional features: Authentication: email and password, Google OAuth, GitHub OAuth. Cloud sync: real time sync with offline queue and retry. Theme system: Default Dark, Blood Moon, Midnight Forest. Keyboard shortcuts: navigation (for example, Ctrl+1–4 for modules), quick actions. Quick capture: global shortcut (for example, Ctrl+K) for instant note or task creation. Import/export: JSON, Markdown, CSV, with optional encryption. Ambient audio: soundscapes and UI sound effects for immersion.

How we built it

Tech Stack

Frontend: React 19.2 + TypeScript 5.9
Build: Vite 6.0
Styling: CSS Modules with custom animations
Backend: Firebase (Auth, Firestore, Cloud Storage)
AI: NVIDIA Nemotron models (via OpenRouter - llama-3.1-nemotron-70b-instruct) Google Gemini API OpenAI (via OpenRouter) AWS SageMaker (for fine-tuned personality models) Hugging Face Einstein API ElevenLabs (text-to-speech voice synthesis)
State: React Context API
Testing: Vitest

Kiro AI Integration

Kiro completely changed the way I build software. The property-based testing gives me confidence that my code truly matches the intent of my specs. The checkpoint system is a lifesaver, letting me roll back to any point in my development session without losing progress. The new Kiro CLI brings my custom agents straight into the terminal, making the entire workflow feel natural and fast. And knowing that Kiro is evolving toward centralized, team-wide management makes it feel not just like a tool, but a platform built for real collaboration.

EVERY feature in this site was created with a spec requirement design phase with documents created by Kiro, which also create story tasks to complete as the feature is being integrated. Some examples of the workflow :

1. Vibe Coding (Exploration Phase) Used natural conversation to explore ideas:

"Make tombstones sway in wind based on priority" → Generated CSS animations with priority-based keyframes
"Add moon phase lighting" → Implemented astronomical calculations and dynamic CSS variables
"Create particle effects" → Built particle system with spawn logic and cleanup

2. Spec-Driven Development (Structured Phase) Created specifications in .kiro/specs/:

interactive-graveyard-physics/ — Requirements, design, and implementation tasks
cursed-typewriter/ — Feature breakdown with user stories and technical design
ghost-writer-ux/ — AI assistant interface specifications

3. Agent Hooks (Automation) Custom hooks in .kiro/hooks/:

auto-test-coverage.kiro.hook — Ensures test coverage for new components
code-quality-analyzer.kiro.hook — Runs ESLint and TypeScript checks
ghost-writer-hooks/ — Creative writing prompts and sentence completion

4. Steering Documents (Consistency) Steering docs in .kiro/steering/:

product.md — Design philosophy and UX principles
tech.md — Technology stack and code style
structure.md — File organization and naming conventions

Development Process

Planning: Created specs for each feature
Implementation: Used vibe coding for rapid prototyping, specs for structured development
Quality: Hooks automated testing and code quality checks
Refinement: Iterated based on feedback and testing
Optimization: Code splitting, lazy loading, performance tuning

Key Implementation Highlights

AI Integration:

Context extraction from writing
API integration with OpenRouter
Fallback patterns for offline mode
Caching for performance

Challenges we ran into

Challenges We Ran Into

Handling complex offline and cloud sync states.
Managing AI provider differences, rate limits, and fallbacks.
Keeping performance acceptable with animations and effects.
Dealing with contentEditable quirks (moved to custom editor).
Balancing strong visual theme with accessibility.

Accomplishments We Are Proud Of

Fully functional multi module productivity suite built under hackathon constraints.

Multi-Agent AI System (Ghost Archive) Built a multi-agent system with historical personalities (Einstein, Cleopatra, Frankenstein) Implemented intelligent agent routing across Hugging Face, AWS SageMaker, and Gemini APIs Added multiple reasoning modes: Chain-of-Thought, Tree-of-Thought, Self-Consistency, and Collaborative Created a workflow engine for complex multi-agent tasks Voice-Powered AI Interactions Integrated ElevenLabs TTS with personality-matched voices Implemented auto-speak with configurable voice settings Built a voice module with browser TTS fallback Advanced Repository Analysis (Mystic Oracle) Deep code analysis with AI-generated insights 8-metric grading system (Activity, Consistency, Productivity, Code Health, Communication, Maintenance, Innovation, Technical Debt)

What We Learned:

During the project, I experimented with Amazon SageMaker to explore training my own custom models for the historical figures. At first, it seemed like the perfect fit — scalable, powerful, and ideal for building personality-tuned AI agents. Some features look great on paper but are not cost-effective unless you have substantial credits, real-time endpoints start at $193+/month, even if barely used.L arge GPU instances are extremely powerful, but not practical without credits . This experience helped me understand where SageMaker shines and when to avoid features that quietly accumulate costs. It also guided my architectural decisions, ultimately leading to a hybrid solution using Hugging Face, Gemini, and local orchestration instead.

Offline-first design pairs perfectly with Kiro’s spec-driven workflow, keeping UX simple while ensuring that core flows are fully defined before coding even begins.

Service-layer architecture made AI integrations far easier to test, mock, and replace, especially when working with Kiro Hooks and modular agent tools.

Kiro works best as a structured development partner: writing clear specs, steering rules, and test cases dramatically improved response quality and reduced iteration time.

Documentation is a force multiplier with Kiro: Specs, steering files, and checkpoints naturally become living documentation, making our repo easier for judges and future collaborators to navigate.