A proactive security platform specializing in detecting and mitigating side-channel attacks on Large Language Models. Utilizes anomaly detection and behavioral analysis to identify malicious queries and protect against data leakage.
A mobile application that extends end-to-end encryption (E2EE) on RCS messaging beyond Apple's initial implementation, ensuring consistent security and privacy across all platforms, including Android and potentially desktop.
A web application that leverages machine learning to predict Steam Deck OLED availability based on historical data, RAM/storage shortage reports, and regional demand. Includes a notification service for users to be alerted when stock is replenished.
AI Persona Guard
A software tool for marketing teams to audit and refine AI-generated advertising content to ensure brand consistency, avoid legal pitfalls, and adhere to ethical AI usage guidelines, specifically addressing the inconsistencies highlighted by Samsung's AI ad usage.
LLM Shield
A security software designed to detect and mitigate side-channel attacks against Large Language Models (LLMs). It dynamically analyzes LLM inference patterns in real-time, identifying anomalies indicative of timing/other side-channel attacks, and adapts parameters to obscure these vulnerabilities.
A revamped podcast player leveraging Apple’s HLS technology, enabling seamless switching between audio and video podcasts from various sources, including offline video download support and enhanced organization features.
A mobile application providing end-to-end encrypted messaging across iOS and Android, building upon Apple's RCS implementation to deliver a privacy-focused communication experience rivaling Telegram or Signal.
A real-time stock tracker and notification service for the Steam Deck OLED, utilizing web scraping and API integration to proactively alert users about restocks across different regions, particularly addressing the reported memory and storage shortages.
A mobile application designed to analyze social media platforms (like Samsung's) for AI-generated content disclosures, providing users with a clear and consistent understanding of when they’re interacting with AI-created material. It uses sentiment analysis to flag potentially deceptive or misleading AI usage.
An e-commerce platform optimizing the sale of refurbished PCs, leveraging real-time component shortage data (especially memory) to dynamically adjust pricing and inventory, as well as providing transparent condition reporting and warranty options, catering to cost-conscious consumers and expanding the accessibility of affordable compute.
A cybersecurity-focused software that builds upon existing password management solutions with advanced features like decentralized access control, adaptive multifactor authentication optimized for remote workforces, and automated data breach risk assessments. The software aims to streamline security within organizations of diverse sizes.
A developer tool that analyzes and explains how language models (specifically Transformers) handle indexical language ('you,' 'I,' 'here,' 'now'). It provides mechanisms to visualize and understand model's internal state during processing of these references to help developers create more reliable and understandable AI systems.
A software platform designed to evaluate the memory and continual learning capabilities of AI agents, specifically focusing on long-term context and adaptive behavior. Leveraging techniques like Linear RNNs and large language models (400B+), the system would assess agents' performance in complex, dynamic environments, providing detailed metrics and insights for AI developers and researchers.
A SaaS platform that analyzes secondhand laptop sales data, identifies pricing trends, and automatically adjusts listing prices on major marketplaces to maximize profit margins. The system reacts to memory chip shortages and component costs, enabling businesses to dynamically optimize their refurbished PC inventory.
A debugging tool for Transformer-based models, specifically designed to analyze and understand how models handle indexical language (e.g., 'you,' 'I'). It identifies and visualizes the model's internal representation of these terms aiding in interpretability and improving contextual accuracy.