Deterministic Retrieval-Grounded Language Models for Clinical Counseling: Large-Scale Multilingual Evaluation with Cryptographically Verifiable Pipelines
Abstract
Large language models (LLMs) show considerable promise for mental health dialogue systems, yet their deployment raises pressing concerns around safety, hallucination, reproducibility, and clinical reliability (Ji et al., 2023; Bommasani et al., 2021). We present a deterministic architecture for AI-assisted counseling that combines retrieval-augmented response generation, structured dialogue management, rule-based risk routing, and a cryptographically verifiable evaluation pipeline. The system was evaluated on two independent datasets spanning 1,895 counseling scenarios in English and Chinese. On 783 English counseling cases, the system achieved mean scores of 4.33/5 for empathy, 3.55/5 for clinical fidelity, and 4.45/5 for safety. On 1,112 Chinese cognitive-behavioral therapy (CBT) scenarios, the corresponding scores were 4.85/5, 4.73/5, and 4.77/5. No system failures or unintended diagnostic outputs were observed across either evaluation. Ablation experiments demonstrate that retrieval grounding and deterministic safety routing each contribute significantly to overall performance, with the former driving clinical fidelity and the latter driving safety. These results suggest that deterministic, retrieval-grounded LLM architectures can serve as a viable foundation for scalable and safe psychological support systems.
Read the paper
Related studies
- Knowledge Management Perspective of Generative Artificial Intelligence
- Navigating Generative Artificial Intelligence Promises and Perils for Knowledge and Creative Work
- Characterizing the evolving research on enterprise content management
- Large language models and multimodal foundation models for precision oncology
- From Generative AI to Extended Reality: Multidisciplinary Perspectives on the Challenges, Opportunities, and Future of Educational Computing
- Generative Artificial Intelligence: A New Engine for Advancing Environmental Science and Engineering
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.