Welcome ! GitHub LinkedIn Email

Hi, I’m Allen Sunny, a master’s student at the University of Maryland, College Park. I build human-centered AI systems and conduct research at the intersection of machine learning, public services, and digital accountability. I’m currently finishing my thesis under the guidance of Professor Ido Sivan-Sevilla.

This site showcases my research, projects, and links to my code repositories.

πŸ” Current Work

πŸ“š Research Projects

Human-Centered AI for the Public Sector

Developing a framework and prototype to support the responsible deployment of AI in public institutions. This work draws on case studies of past failures, emphasizes human-centered design, and proposes a roadmap for post-deployment evaluation and accountability.

  • πŸ“„ Thesis (in progress)
  • πŸ’» Prototype coming soon

Trust in Transparency: How Explainable AI Shapes User Perceptions

A qualitative study based on semi-structured interviews that explores how users emotionally respond to AI-generated explanations. The findings highlight that trust is shaped not just by transparency, but also by reliability, clarity, and perceived control.
- πŸ“„ Paper

Explainability and Trust in AI: A Quantitative Approach

A survey based study that evaluates how different types of AI explanations influence user perceptions of fairness, trust, and decision legitimacy.
- πŸ“„ Paper

Oxford AI Monitoring System

A lightweight, automated web scraper designed to track corporate adoption of foundation models. This tool supports transparency and governance efforts by mapping real-time deployment trends across public-facing websites.
- πŸ’» Code

πŸ› οΈ Open Source Tools

βš™οΈ Applied Systems

Dark Pattern Analysis

A browser extension that detects deceptive UX elements using DOM analysis and LLM-based classification.
- πŸ“„ Paper
- πŸ’» Code

RADAR: Retrieval-Augmented Data Analysis and Representation

A RAG-powered system that turns natural language into explainable data visualizations using LLaMA 3.1 8B.
- πŸ“„ Paper
- πŸ’» Code

πŸ“¦ R Tools for Interpretable Machine Learning

TangledFeatures

Performs feature selection in highly correlated spaces for use with interpretable models.
- 🌐 Package Site
- πŸ“„ Documentation

StructuralDecompose

Decomposes time series into components while remaining robust to level shifts.
- 🌐 Package Site
- πŸ“„ Documentation

🌐 About This Site

Built with Quarto
Hosted on GitHub Pages
Always evolving.