Protopia Roundtrip Protection is a data privacy solution for LLM inference workflows that prevents plaintext exposure of sensitive data throughout the entire AI inference lifecycle — from prompt input to model output. The core technology is the Stained Glass Transform (SGT), a proprietary stochastic (randomized) transformation algorithm that converts sensitive data into a protected representation before it leaves the client environment. This transformed data is sent to the hosted LLM inference endpoint, processed without ever existing in plaintext on the hosting infrastructure, and the model output is returned in a form that only the authorized client can decrypt using a private output key. Key characteristics of the approach: - The LLM and its hosting infrastructure never receive or process plaintext data - Only the end-user client holds the keys to reconstruct full plaintext inputs and outputs - Designed for multi-tenant managed inference environments where the client does not control the underlying compute The solution targets regulated industries (e.g., healthcare) and enterprises using third-party managed LLM inference platforms. It is positioned as an alternative to data masking, entity redaction, firewall-based blocking, and on-premises deployment — approaches that either degrade model performance or limit scalability. Protopia has partnered with Lambda (Lambda Cloud) to deliver this capability on Lambda's managed GPU inference endpoints, enabling enterprises to use hosted LLMs while maintaining data confidentiality end-to-end.