Skip to main content

​
Creator

Kory Smith

Role: Creator & Lead DeveloperContributions:
  • Full-stack development
  • Machine learning implementation
  • API architecture & design
  • SDKs (Python & JavaScript)
  • Documentation & deployment

​
Project Background

Guardian API evolved from an academic machine learning project focused on detecting sexism in social media text. What started as a binary classification problem for a CSC 380 class project has grown into a comprehensive, production-ready content moderation system.

​
Academic Foundation (Fall 2025)

Course: CSC 380 - Machine Learning Institution: University of Arizona Original Goal: Build a sexism classifier using LASSO regression Initial Scope:
  • Train binary classifier on ~40,000 labeled tweets
  • Achieve >80% F1 score
  • Demonstrate practical ML engineering

​
Evolution to Production

After successfully completing the academic requirements, I chose to expand the project into a full-fledged API service: Phase 1: Multi-Model Architecture
  • Added toxicity transformer model
  • Integrated rule-based heuristics
  • Designed ensemble aggregation system
Phase 2: Production Infrastructure
  • Built FastAPI backend
  • Implemented rate limiting and caching
  • Added comprehensive error handling and logging
Phase 3: Developer Experience
  • Created Python and JavaScript SDKs
  • Generated comprehensive documentation
  • Set up interactive API playground
Phase 4: Advanced Features
  • GPU acceleration support
  • Batch processing capabilities
  • Custom deployment options

​
Philosophy

Guardian API is built on several core principles:
Open Source & Explainable
  • Full source code available on GitHub
  • Clear model explanations
  • Documented decision-making process
  • No black-box algorithms
Multi-Model Validation
  • Never rely on a single model
  • Ensemble voting reduces false positives
  • Continuous evaluation and improvement
  • Academic rigor meets production standards
Configurable & Extensible
  • Adjustable thresholds
  • Custom rule sets
  • Pluggable model architecture
  • Self-hosted or cloud options
Built for Engineers
  • Clean REST API
  • Type-safe SDKs
  • Comprehensive documentation
  • Interactive examples

​
Technology Stack

Guardian API leverages modern tools and frameworks:

​
Backend

  • FastAPI: High-performance async Python web framework
  • PyTorch: Deep learning framework for transformer models
  • Scikit-learn: Machine learning for LASSO classifier
  • Pydantic: Data validation and settings management

​
Models

  • Custom LASSO: Binary sexism classifier
  • HuggingFace Transformers: Pre-trained toxicity detection
  • NLTK: Natural language preprocessing
  • Custom Rules: Regex-based pattern matching

​
Infrastructure

  • Uvicorn: ASGI server
  • Redis/Upstash: Rate limiting and caching
  • Docker: Containerization
  • GitHub Actions: CI/CD (planned)

​
SDKs

  • Python: Type hints, async support
  • TypeScript: Full type definitions
  • Requests/Fetch: HTTP clients

​
Recognition

Academic Excellence: Guardian API received high marks in CSC 380 for demonstrating production-grade ML engineering, going well beyond the standard academic project scope.
Open Source: The project is fully open source and welcomes contributions from the community. See our GitHub repository for contribution guidelines.

​
Contact

​
Acknowledgments

Special thanks to:
  • University of Arizona - For providing the academic environment and resources
  • CSC 380 Course Staff - For guidance on ML best practices
  • HuggingFace - For pre-trained transformer models
  • FastAPI Community - For excellent web framework
  • Open Source Community - For the tools that made this possible

Want to Contribute?

Guardian API is open source and welcomes contributions. Whether it’s code, documentation, or ideas, contributions are appreciated.