MindFrame
An open-source research platform for delivering structured psychological interventions through conversational AI.
Interventions as directed graphs.
Therapeutic conversations are represented as directed graphs rather than linear scripts. Each step defines prompts, evaluations, and transition conditions.
This structure allows conversations to branch and adapt based on participant responses.
Adaptable to local contexts.
The same intervention structure can be configured with different therapeutic goals, cultural considerations, and referral pathways depending on where it is deployed.
Local services, population characteristics, and contextual factors can be specified without modifying the underlying intervention logic.
Transparent and auditable.
All prompts, judgements, and transition conditions are defined in plain-text templates. The logic governing each conversation can be inspected and reproduced.
This supports the accountability and reproducibility requirements of research contexts.
Structured data collection.
Judgements run during conversations to extract structured information from natural language. Notes record observations, scores, and summaries as data.
Dashboards and data exports provide access for researchers and supervisors.
Safety monitoring.
Risk flags passively monitor conversation data. Interruptions can pause conversations and redirect to appropriate steps when specified conditions are met.
Human oversight mechanisms allow supervisors to review flagged conversations.
Evidence retrieval.
Responses can draw on curated evidence databases through semantic search. Clinical materials, treatment manuals, and guidelines can be annotated and indexed.
Retrieval-augmented generation incorporates relevant content into conversation turns.