MICA: Motivational Interviewing Conversational Agent

This is a REDCap External Module (EM) developed to support the MICA Project — an AI-powered Motivational Interviewing (MI) chatbot for engaging participants in structured conversations about substance use and behavior change.

---

Overview

MICA is designed as a longitudinal digital intervention, embedded in REDCap, with the following goals:

• Support behavior change through MI-aligned chatbot conversations
• Operate securely within Stanford's infrastructure
• Provide flexible follow-up sessions with context-aware chat histories
• Automatically handle scheduling, session flow, logging, and post-survey logic

MICA interacts with Stanford’s SecureChatAI instance, ensuring all data remains compliant with PHI restrictions (site-to-site VPN, no external API calls).

---

Key Components

This EM (MICA)

• Installs a standalone chatbot interface inside REDCap
• Handles:
  • Participant session detection and routing
  • Dynamic System Prompt configuration per session
  • Session transcript capture and saving to REDCap fields
  • Session catch-up and summarization logic
  • Chatbot context management (e.g., system/assistant/user roles)
  • Completion logic to forward users to post-surveys
• Extends functionality of SecureChatAI EM (required dependency)

Chat Frontend

• Built in React; produces a static bundle injected into the EM view
• Resides in /static/ and renders within REDCap
• Can optionally be embedded elsewhere

---

Installation

1. Clone this repo into your REDCap /modules/ directory:

   git clone https://github.com/YOUR_ORG/mica-em.git mica_vX.X.X

2. Enable the module only on the REDCap project used for MICA

3. Ensure SecureChatAI EM is installed and configured

---

Required REDCap Project Setup

Longitudinal enabled:
Must have events such as baseline_arm_1, session_2_arm_1, ..., session_7_arm_1

Instruments:

• session_info – stores raw chat logs and timestamps
• posttest – post-session survey
• month3_fu – optional final follow-up survey

Fields:

• raw_chat_logs
• session_timestamp
• session_info_complete
• des_mica – used for opt-in/out decisions

---

Session Flow

• On login, participant is routed to their eligible session
• If prior sessions exist, a summary is injected into the system prompt
• Chat session completes, logs saved to session_info
• Participant is redirected to posttest, and if applicable, to month3_fu

---

Project Settings

• chatbot_system_context_general – default global system prompt
• chatbot_system_context_session_X – optional session-specific overrides
• chatbot_end_session_url_override – optional redirect override for post-survey

---

Security Notes

• All chat processing uses Stanford’s SecureChatAI instance
• No third-party API calls (e.g., OpenAI, Claude, Gemini)
• All data remains within Stanford’s secure REDCap and hosting environment

---

Dependencies

• SecureChatAI EM
• REDCap v13+
• PHP 7.4+
• Node.js for React frontend rebuild

---

Dev Notes

To rebuild the React frontend:

cd frontend
npm install
npm run build