A Hybrid Architecture for Responsible AI

RIACT (Record, Insight, Analyze, Coach, Track) addresses the problem of retrospective burnout detection by providing students with structured visibility into their study habits. The system avoids the common pitfall of using opaque AI models for health diagnostics by employing a hybrid architecture:

• Deterministic Rule-Based Detection: Burnout signals are identified using transparent, week-over-week behavioral comparisons. By relying on hard-coded rules rather than model inference, the system ensures that alerts are auditable and explainable.
• Constrained LLM Coaching: Large Language Models are used only to contextualize patterns and generate recommendations. To prevent hallucination or over-diagnosis, the LLM is strictly constrained to a fixed output schema, ensuring the AI acts as a supportive coach rather than a medical authority.

Design Principles for User Agency

The system is built on the premise that productivity tools often record data without providing actionable interpretation. RIACT shifts this by framing all AI output as an "observation" rather than a "diagnosis." This design choice is critical for maintaining user trust and preventing the anxiety that can arise from automated health labeling. Furthermore, the system limits data collection to self-logged behavioral fields, ensuring that the student remains the primary owner and curator of their study data.