CLARA

An agentic analytics system that extracts semantic representations from multi-party conversations as shared artifacts to support agent retrieval and reasoning.

Paper: CLARA: An AI-Augmented Analytics Dashboard for Collaboration Literacy (AIED 2026) · arXiv

System Workflow. (A) CLARA transcribes discussion in real time, computes psycholinguistic metrics, and invokes the LLM to produce collaboration assessments and concept maps. (B) Transcripts and artifacts are embedded and indexed in distinct database collections. (C) Users can interact with LLM-produced artifacts through the dashboard; (D) The CLARA Agent reasons and iteratively calls tools to query different artifacts for synthesizing responses; (E) The interface where users chat with the agent.
The user interacts through chat and an editable dashboard; every edit is persisted to an audit log. Within a turn, the agent works from the query, a memory module that holds a running model of the user, and artifacts retrieved with field-level authorship. An asynchronous reflection pass updates the memory for the next turn.
An earlier iteration of the dashboard (BLINC), showing the psycholinguistic metrics and the Discussion Pulse rolling summaries.
The current CLARA interface.