rhemix-cs

Rhemix is a Python-based ecosystem designed for generating synthetic, stateful dialogue datasets, with a focus on supporting evaluation and research in conversational AI. The platform enables users to create multi-party conversations by combining structured inputs such as participant profiles, scenarios, domain constraints, and discourse-state operations. Its core aim is to produce conversations that are generated rather than scripted, providing inspectable structure and metadata for each dialogue instance.

The system is particularly intended for evaluation engineers, data scientists, and researchers who require controlled dialogue data for tasks like conversational AI evaluation, compliance testing, regression tests, and research that benefits from access to both conversational and mental state modeling. Unlike datasets sourced from scraped or hand-authored conversations, Rhemix emphasizes the creation of dialogues with explicit state tracking and quality checks, making it suitable for workflows that demand transparency and reproducibility in dialogue generation.

Rhemix is organized as an open framework with distinct components. The rhemix-engine provides the discourse runtime, a dashboard, and batch generation capabilities, while rhemix-profiles handles the infrastructure for generating participant profiles. Domain packs define the specific world for each dataset, including roles, prompts, tools, constraints, and evaluation rules. The first available domain pack, distributed under a private license, is tailored for customer-service dialogues and includes synthetic profiles, personally identifiable information (PII), transfer scenarios, channel artifacts, and quality checks.

0 license, while the domain packs have varying licenses depending on their content. The system’s approach centers on combining domain-specific ingredients into new, inspectable conversations, offering a controlled alternative to template-based or reused dialogue sources.