The Sentiment Data Analysis is a process that uses AI to analyze how students feel when engaging with the Companion Bot in the Progressive Web App (PWA) and Companion Native App. By performing automated sentiment analysis, the system identifies whether interactions are positive, neutral, or negative, providing an evidence-based understanding of how effectively the bot supports students.

At this stage, all analysis focuses exclusively on the Companion PWA and other data sources. Analysis of the Companion Native app will begin once sufficient data is available.

Two key sentiment reports are planned:

• Student–Bot Interactions (PWA) – measuring students’ emotional responses when using the bot.

• Mentor–Student Interactions – assessing conversations with mentors to monitor satisfaction and engagement.

The primary objective is to determine whether students feel comfortable, supported, and satisfied when using the Companion Bot in the PWA. The project also aims to:

• Evaluate the emotional tone of interactions (positive, neutral, negative).

• Identify the most frequently searched topics and questions.

• Assess the level of trust and confidence students place in the bot to meet their needs.

• Generate clear, actionable insights to guide design improvements and feature development in future iterations.

Ultimately, the project seeks to create a more effective, trustworthy, and engaging student experience.

Sentiment analysis uses AI to automatically detect the emotional tone of written text. In this process, all data is gathered passively, without requiring any additional input from students.

All activity within the Companion PWA—such as logins, selected options, completed tasks, and search queries—is automatically recorded. At the same time, mentors document their interactions with students through a centralized system that includes messages exchanged via platforms like WhatsApp or email.

Data collection and analysis are organized as follows:

1) Student–Bot Interactions (Companion PWA):

• All interactions—logins, menu selections, completed tasks, and search queries—are automatically recorded by the system.

• No surveys or manual inputs are required.

2) Mentor–Student Interactions:

• Conversations are documented manually by mentors through a shared system (e.g., logs from WhatsApp or email).

The collected data is then analyzed using AI models designed to perform specific analytical tasks, such as:

• Identifying whether the sender is a student or a mentor

• Classifying the source of the interaction (e.g., Companion Bot, WhatsApp, email)

• Assessing the sentiment (positive, neutral, or negative)

• Generating reports to support decision-making and improvements

This approach enables continuous, non-intrusive monitoring of the student experience, allowing teams to make informed decisions and enhance support services without disrupting the learning process.

The Sentiment Data Analysis for the Companion PWA has moved through a structured, evidence-driven workflow. Each data set (Student ↔ Bot, Student ↔ Mentor, and—later—Student ↔ Support Agent) follows the same three-phase cycle: AI Tagging → Data Preparation → Reporting & Visualisation. 

We will continue to simulate chat bot conversations before going live, simulating 100,000 Spanish speaking students and share the results, run the same for Portuguese, French, etc. and continue to pivot accordingly. 

Guardrails AI Introduces Snowglobe: The Simulation Engine for AI Agents and Chatbots - MarkTechPost