A Conversation-Led, Agent-Orchestrated Architecture with Persistent and Temporal Memory for Mental-Health Assessment Support
Year: 2026
Type: article
Abstract: MENTIS is a conversation-led mental-health monitoring and assessment-support research system. This paper describes its service-oriented architecture: a Next.js frontend, a FastAPI application boundary, PostgreSQL for authoritative records, Redis for workflow-continuation state, Neo4j with Graphiti for temporal knowledge-graph memory, and an n8n planner/executor workflow that is resumed across user turns. We present the relational data model, a turn-lifecycle protocol that separates durable conversation records from ephemeral workflow state, the agent topology, and a nine-family input specification covering provenance, missingness and synthetic data. The design is grounded in a 67-entry research register. We report implementation status candidly and claim no diagnostic accuracy, clinical benefit or retrieval-speed advantage. Instead, we specify a staged evaluation protocol to be completed before any such claim. MENTIS is a research prototype and is not a clinical tool.
Authors: Jash Nankani, Jai Saraswat
Language: English
Subfield: Applied Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7233847951
