QuASy Framework: AI-Powered IP-Safe Intelligence

Propensic introduces the QuASy Framework, a cutting-edge AI system designed to protect intellectual property while leveraging the power of public and private language models. QuASy enables secure, anonymized querying of public LLMs by abstracting sensitive data—like trade secrets, patents, and proprietary formulas—using a our own patented (pending) anonymization engine powered by spaCy and regex.

🧠 Private Language Model (PLM) Tuning with Feedback Loops

Every anonymized query sent to public LLMs is tracked, and the delta between public and private responses is used to continuously tune a secure, domain-specific PLM. This ensures your proprietary knowledge evolves safely and intelligently.

🛠️ Developer Tools & Interfaces

• CLI, Web UI & API: Easily anonymize queries and simulate LLM responses via intuitive command-line tools and interactive web or REST APIs.
• PLM Configuration: YAML-based setup using vLLM and Hugging Face Transformers, supporting quantization and acceleration.

🔍 IP Detection & Abstraction

Advanced logic detects and replaces sensitive entities with placeholders, maintaining secure mappings for PLM reconstruction—ensuring your data remains protected while still benefiting from public AI capabilities.

📦 Architecture

Built with a modular design:

• feedback compliance
• delta tuning
• IP governance

📄 Patent Pending Innovation (63/914,079)

QuASy is backed by a provisional patent assessment under 35 U.S.C., highlighting novel features like feedback-driven PLM tuning and IP-aware abstraction rules.

📊 Visuals & Documentation

Includes UML diagrams, ASCII architecture maps, and a Jupyter-based PoC notebook for transparency, onboarding, and technical clarity.