The Science Behind EQ

EQ is not a wellness app. It is a clinical-grade AI system grounded in established psychological frameworks, computational linguistics, and modern large language model research. This page explains the scientific foundations that make EQ work.

Emotional Granularity and the Limits of Mood Tracking

Most mental health apps ask patients to rate their mood on a 1–10 scale. This approach has a fundamental problem: it collapses the full dimensionality of human emotional experience into a single number. A patient who rates their mood a 4 might be experiencing guilt, grief, or fear — three conditions that call for entirely different therapeutic responses.

EQ is designed around the concept of emotional granularity — the ability to distinguish between discrete emotional states rather than treating all negative affect as equivalent. Research by Lisa Feldman Barrett and others has shown that people with higher emotional granularity demonstrate better emotional regulation, lower rates of depression relapse, and more adaptive coping strategies. EQ operationalizes this insight computationally.

The Emotional Extraction Pipeline

When a patient sends a message to EQ, the text is processed through a structured pipeline built on LangChain and OpenAI’s large language models. The pipeline performs multi-label emotion classification — identifying not just whether an emotion is present, but which specific emotions are present, how intensely they are expressed, and how they relate to one another within the message.

The taxonomy EQ uses draws from established categorical emotion models, including basic emotion theory (Ekman) and more granular appraisal-based frameworks. Emotional labels extracted by EQ include but are not limited to: Guilt, Gratitude, Pride, Anxiety, Grief, Shame, Hope, Anger, Loneliness, and Fear. Each label is accompanied by an intensity score derived from linguistic cues in the message.

Longitudinal Emotional Profiling

A single emotional data point is not clinically useful in isolation. EQ’s value comes from the accumulation of data over time. As a patient continues to interact with the AI between sessions, EQ builds a longitudinal emotional profile: a time-series record of which emotions appear, at what intensity, and in what patterns.

This longitudinal data enables the detection of:

  • Persistent emotional themes (e.g., recurring guilt across multiple weeks)
  • Emotional trajectories (e.g., steady decline in expressed anxiety over a treatment course)
  • Acute shifts (e.g., a sudden spike in grief or hopelessness between sessions)
  • Emotional complexity changes (e.g., a patient whose emotional vocabulary becomes more nuanced over time, indicating therapeutic progress)

Session Prep Prompt Generation

EQ’s session prep prompts are not templated summaries. They are generated dynamically using the full context of a patient’s recent emotional profile. The generation model is instructed to produce a clinically-oriented brief that:

  • Surfaces the most clinically salient emotional themes from recent conversations
  • Flags any notable shifts or acute emotional signals since the last session
  • Suggests specific areas the therapist may want to explore based on the patient’s expressed emotional state
  • Is written in language appropriate for a licensed clinician, not a consumer product

This means every session brief is specific to one patient at one moment in time. It is not a generic summary — it is a structured clinical document generated from real patient-generated data.

Why SMS Works for Between-Session Data Collection

The choice of SMS as the patient interface is not arbitrary. Research on ecological momentary assessment (EMA) — the practice of collecting real-time data from patients in their natural environment — consistently shows that SMS-based data collection achieves significantly higher compliance rates than app-based or web-based alternatives. Patients respond to text messages. They do not open apps.

EMA has been validated as a method for capturing emotional data that is more ecologically valid than retrospective self-report (i.e., what a patient tells their therapist in a session about how they felt during the previous week). Memory is reconstructive. SMS responses captured in the moment are closer to ground truth.

The Role of Conversational AI

EQ does not use a structured questionnaire format. Patients are not asked to rate their mood or answer checkbox questions. Instead, they engage in open-ended natural language conversation with an AI that is trained to be warm, non-judgmental, and therapeutically informed.

This matters because structured instruments often fail to capture the emotional nuance present in free expression. A patient who rates their anxiety as a 3/10 may still, in natural language, express language patterns consistent with significant distress. EQ’s NLP pipeline captures what structured instruments miss.

What EQ Is Not

EQ is a clinical decision support tool. It is not a replacement for therapy, a diagnostic instrument, or a crisis intervention system. The emotional data EQ provides is intended to augment the clinical judgment of a licensed therapist — not to replace it. EQ helps therapists arrive more informed. What happens in the room is still entirely the domain of the clinician.