AI Processing Details

Last Updated: March 2026 Version: 2.0

This page explains how Versa uses AI to process your data — what models are involved, what they do with your information, and how decisions are made. This information is provided pursuant to Articles 13, 14, and 15(1)(h) GDPR.


How AI Is Used in the Versa Platform

Versa uses AI for three purposes:

1. Realtime Training Conversations

When you practice a training scenario, you speak with an AI persona that responds in real time. The AI receives your speech (converted to text) and generates responses based on the scenario instructions your trainer has set up.

What the AI sees: Your speech (as text), the scenario description, the AI persona's role and instructions, and the conversation history within that session.

What the AI does NOT see: Your name, email, account details, or data from other users' sessions.

Providers (your organisation or you select which to use):

ProviderLocationHow data is handled
Microsoft Azure (GPT Realtime)EU (Sweden Central)Data processed in EU. Not used for Microsoft's model training.
Google Gemini LiveEU (configurable)Data processed in EU. Not used for Google's model training.
Hume AI (EVI)USAAudio processed for emotion-aware responses. Not stored after processing. Not used for Hume's model training. Hume derives vocal characteristics (tone, pace, pitch) but does not perform speaker identification or voiceprint extraction.
xAI (Grok)USAText processed for conversation. Deleted within 30 days. Not used for xAI's model training (User Content). xAI may create de-identified usage derivatives.

For US-based providers (Hume, xAI), data transfers are protected by Standard Contractual Clauses. Your organisation can choose which providers are available.

2. AI Feedback

After a training session, Versa's AI reviews your performance and generates feedback. This is the core of the training experience.

How it works:

  1. The AI reads your conversation transcript and the feedback criteria set by your trainer — e.g., "Did the trainee show empathy?", "Was the explanation clear?".
  2. The AI scores your performance against each criterion.
  3. The AI generates coaching recommendations — specific, actionable suggestions for improvement.

What the feedback is based on:

  • Your conversation transcript (what you said)
  • The scenario's feedback criteria (what "good" looks like)
  • The AI persona's instructions (the context of the conversation)

What the feedback is NOT based on:

  • Your identity, demographics, or personal characteristics
  • Other trainees' performance
  • Data from outside this specific training session

The feedback model: Google Vertex AI (Gemini 2.5 Pro), processed in the EU (europe-west1). Google does not use your data to train its own models.

Important: AI feedback is educational guidance, not an automated decision. Scores and feedback are tools for learning, not final judgments. Your trainer reviews AI feedback and can correct inaccuracies. If your organisation uses feedback scores for formal assessments (grading, employment decisions), your organisation is responsible for ensuring appropriate human review and safeguards under Article 22 GDPR.

3. AI-Assisted Scenario Creation

Trainers can chat with an AI assistant (Vi) to build training scenarios. Vi helps structure the scenario, define feedback criteria, and configure AI personas.

What Vi sees: The trainer's instructions and any content the trainer provides (including URLs fetched via Jina AI).

The model: Google Vertex AI (Gemini 2.5 Flash), processed in the EU (europe-west1).


How AI Feedback Scores Are Generated

Because AI feedback affects your training experience, here is a more detailed explanation of how scores are produced:

Step 1 — Input assembly. The system assembles your conversation transcript, the feedback criteria (titles and descriptions), and the scenario context (what the AI persona was simulating).

Step 2 — Criterion-by-criterion assessment. For each feedback criterion (e.g., "Active Listening", "Clear Communication"), the AI reads the relevant parts of your conversation and assesses whether the criterion was met, partially met, or not met. The AI produces a score and a text explanation for each criterion.

Step 3 — Coaching generation. Based on the criterion scores and the conversation content, the AI generates specific coaching recommendations — what you did well and what to practice.

Step 4 — Output. The scores, explanations, and coaching recommendations are presented to you and your trainer. The full conversation transcript is available alongside the feedback so you can verify the AI's reasoning.

Limitations:

  • AI feedback reflects patterns learned from large-scale language training, not domain expertise. It may miss context that a human expert would catch.
  • Scores are not deterministic — the same conversation reviewed twice may produce slightly different scores.
  • The AI does not "understand" your situation. It assesses communication patterns, not intentions or emotions.

Your Rights Regarding AI Processing

Access: You can view all AI-generated feedback, scores, and coaching recommendations in your session history. The underlying conversation transcript is always available.

Explanation: This page provides information about the logic involved in AI feedback. If you need further details about how a specific score was generated, contact your trainer or privacy@versa.training.

Object: You can object to the use of your data for AI model improvement at any time by contacting privacy@versa.training. Opting out does not affect your access to the platform or the quality of your training experience.

Erasure: You can delete individual sessions (including transcripts and feedback) at any time. Account deletion removes all your data within 30 days.

Human review: If AI feedback is used for a decision that significantly affects you (academic grading, employment evaluation), you have the right to request human review of that decision from your organisation.


Data Used for AI Improvement

When you provide feedback on your training experience (ratings, thumbs up/down), this feedback — along with the associated conversation context — may be used to improve Versa's AI models. This is explained in detail in our Privacy Policy (Section 5).

Key protections:

  • Your data is pseudonymised before any training use (names, emails, and identifying details removed)
  • Identifying concepts in transcripts are replaced with generic equivalents before training use
  • You can opt out at any time without service impact
  • Data from opted-out users is never used for training
  • Our AI providers do not use your data for their own model training

For organisations (B2B): the terms of AI data use are governed by the Data Processing Addendum (DPA) and the feedback licence in the Terms of Service. All data use mechanisms are negotiable.


Sub-Processors Involved in AI Processing

ProviderRoleLocationTraining policy
Google (Vertex AI)Feedback, scenario creation, realtimeEUDoes not train on customer data
Microsoft (Azure)Realtime conversationEUDoes not train on customer data
Hume AIEmotion-aware realtime conversationUSA (SCCs)Does not train on API data
xAI (Grok)Realtime conversationUSA (SCCs)Does not train on User Content; may use de-identified derivatives
SupabaseStorage of transcripts and feedbackEUDoes not train on customer data

Full sub-processor list: versa.training/legal/subprocessors


Questions about AI processing? Contact privacy@versa.training