S

self-learning

HelmOS is a modular, local-first operating layer that wraps existing AI agents with intent verification, human governance, trust management, workflow orchestration, auditing, and controlled self-improvement. Designed to keep humans at the center of decision-making, HelmOS ensures that actions remain explainable, accountable, and aligned with user intent while providing a standardized framework for building, managing, and governing intelligent agent systems. https://roxanneardary.com/helmos/

Project SynthesisMotion is an open-source AGPL-3.0+ research framework for adaptive motion intelligence systems. It combines real-time vehicle identity modeling, physics-informed dynamics, and adaptive control with a formal safety verification layer. The system is designed to continuously learn how a physical vehicle behaves, adapt to changes in real time, and enforce strict safety constraints during operation. https://roxanneardary.com/project-synthesismotion/

ChameleonAI is an open-source, hardware-agnostic AI inference optimization platform designed to deliver maximum performance, energy efficiency, and adaptability across any AI hardware, including Nvidia, AMD, Intel, Huawei, Apple, and emerging accelerators.

It features self-learning optimizations, predictive scheduling, mixed-precision tuning, and security-aware execution, ensuring models run faster, smarter, and safer. ChameleonAI supports cloud, edge, and embedded deployments, with developer-friendly profiling, benchmarking, and plugin tools for full control and transparency. https://roxanneardary.com/chameleonai/

LexiTown is an open source, real-time, multilingual chat platform with adaptive AI that learns community-specific slang, shorthand, and code. Designed as a modern digital town square, it fosters safe, inclusive, and engaging conversations across communities worldwide. The AI continuously self-improves, translating accurately while preserving each community’s unique voice.

A playful, trauma-aware emotional intelligence web project designed especially for neurodivergent minds.

AZ-Lite is a compact, AlphaZero-inspired chess engine written in Python. It combines Monte Carlo Tree Search (MCTS) with a lightweight policy-and-value neural network, learning entirely through some self-play.

[UOC] Practices from the course Software Design Patterns

[UOC] A repo containing links to all my other college projects

[UOC] Challenges from the UOC Enterpreneurship Course

[UOC] Web Standards and Languages HTML, CSS practices

[UOC] A repo containing the various practices I'll work on in Human-Computer Interaction

[UOC] A repo to host my practices for the Operating Systems course