RightTouch Launches 'Conversation Harness' for AI Self-Improvement and Operational Safety

Overview RightTouch (Headquarters: Shinagawa-ku, Tokyo; CEOs: Shuhei Nomura / Daito Nagasaki), a provider of AI contact center infrastructure for enterprises, has announced the launch of 'Conversation Harness,' a new feature for the QANT AI operator that enables operational quality self-improvement and ensures safety.

Conversation Harness serves as a foundation for continuously enhancing the quality of AI operator operations without relying solely on manual labor. The AI automatically analyzes response logs, automating everything from issue detection to the creation, validation, and implementation of improvement proposals, thereby accelerating the knowledge self-improvement cycle. By designing the system so that humans only handle final decisions on implementing improvements, the system achieves both reduced operational workload and quality assurance.

Furthermore, as a final risk hedge against critical accidents during interactions—such as hallucinations—the system is equipped with multi-layered guardrail functionality.

By functioning through both the 'self-improvement cycle' and 'multi-layered guardrails,' the quality of the AI operator improves the more it is operated, while critical accidents are intercepted.

Background of Development While the adoption of generative AI in contact centers is spreading rapidly, maintaining operational quality has become a challenge. Massive daily tasks—such as visual inspection of AI operator logs, continuous adjustments to prompts, and knowledge updates—make it structurally difficult to keep up with manual labor alone.

Furthermore, there is the risk of hallucinations inherent to generative AI. Conversation Harness addresses these issues by combining an automated self-improvement cycle with real-time multi-layered guardrails.

Components of Conversation Harness - Post-interaction self-improvement cycle (Offensive): Analyzes interaction logs to continuously improve knowledge and prompts. - Multi-layered guardrails during interaction (Defensive): Instantly blocks critical accidents such as hallucinations.