Skip to main content

The platform behind every PokaMind build

PokaMind structures yours into one interaction model and runs it as roleplay practice, drills and a team view on a screen and, in pilot, on a robot. Data is only ready for AI in relation to a specific use, which is why every build starts from a scoped situation.

interaction-ready data
nounInteraction-ready data is an organization's knowledge and conversation signals, structured for one specific human-facing use, so that an AI can act on it and a person can check it.
  1. People

    What it holds

    What it holds:

    Who is on each side of the conversation, what they want, and five dials for how hard they push.

    Built from

    Built from:

    Your roles, personas and customer types.

  2. Knowledge

    What it holds

    What it holds:

    What your documents say, split into passages that keep their source.

    Built from

    Built from:

    Playbooks, policies, handbooks, product documents.

  3. Moments

    What it holds

    What it holds:

    The situations, how they open, how they turn, and the techniques that work.

    Built from

    Built from:

    Your training material and the cases your team reports.

  4. Signals

    What it holds

    What it holds:

    What can be observed in practice: words, voice, face and body, described with 31 named communication styles.

    Built from

    Built from:

    The practice sessions your people take part in.

Checked by code. Authored layers approved by a person.

Diagram. The four layers of a PokaMind interaction model.

What goes in

  • 100,000characters per document
  • 1 to 10modules from one document
  • 9languages
  • 5dials for the other person
Documents
PDF, DOCX, PPTX, TXT and MD, up to 100,000 characters each.
From one document
A planned course of 1 to 10 modules, with every idea in the document covered exactly once.
People and roles
Who is on each side, what they want, and five dials for how hard the other person pushes.
Languages
Nine: English, Arabic, Spanish, French, Italian, German, Catalan, Chinese and Swedish, with Arabic authored in Egyptian or Saudi dialect.
Workshops
Where the knowledge lives in people rather than files, we structure it with them.
What we do not need
Customer records, call recordings or CRM data.

What a practice session records

Four channels of observable communication, and 31 named communication styles calibrated against the person's own baseline in that session. The analysis runs while they speak, and the written breakdown arrives when they finish.

Words

From the transcript

  • Clarity
  • Questions
  • Reasoning
  • Other-focused wording
  • Word variety
Voice

From the audio

  • Speaking pace
  • Fluency
  • Pitch variety
  • Steadiness
  • Projection
Face

From video frames analyzed on PokaMind's servers in the EU

  • Eye contact
  • Expression range
  • Visible reactions
  • Brow and jaw movement
Body

From posture, in the same video frames

  • Posture
  • Open stance
  • Movement
  • Shoulder and hand movement
  • Head movement
What it is designed not to infer
  • Emotions or mood
  • Stress or nervousness
  • Personality
  • Honesty, intent or attitude
  • Who someone is: no face or voice recognition

31 named communication styles, such as direct, measured, acknowledging, pausing and investigative. Every one describes behavior.

  • A signal that was not captured is shown as not captured. It is never filled in with an average.
  • Camera frames are analyzed on PokaMind's servers in the EU, not on the learner's device. Speech transcription and language analysis run through AI sub-processors listed in the data processing terms we provide during evaluation.
  • It is designed not to infer emotions, inner states or personality, and it never identifies anyone from their face or voice.

What is kept, and who sees it

  • Measured

    Computed from the signal: the channel measures, the communication-style profile and a time series across the session.

  • Interpreted

    Written by a language model from the transcript together with the measured signals: scores on areas such as clarity, presence, reasoning, technique application and adaptability, and the written breakdown. Stored apart from what was measured.

  • Missing

    A channel that could not be measured is recorded as missing and shown as not captured, never filled in.

Also kept with the record: the session transcript and the audio recording, under the data processing terms we provide during evaluation.

  • The person who practiced: their full breakdown and history.
  • Managers, in PokaMind: each person's practice activity and session scores over time, and a team report written by AI from those scores, with names replaced by numbered labels. Never the conversation, the transcript or the breakdown.
  • On the robot: session records under a code, never a name.

Five rules every build follows

  1. Observable only

    Channel measures and the 31 style labels describe behavior. A few older labels inside the app are still being renamed to match.

  2. What was measured is kept apart from what was interpreted

    Measured values, model judgments and robot behavior decisions are separate fields, so a reviewer can tell which is which.

  3. A missing signal is recorded as missing

    Nothing absent becomes a zero. On the robot, a channel that did not report is not published at all.

  4. The model proposes, code checks, a person decides

    Course plans pass a coverage validator, scenarios pass a readiness check, briefing quotes are checked against their source, and a person approves before anything authored reaches your people.

  5. Pseudonymous where data travels

    On the robot and in its analytics export, people are codes, not names, and speech text is off on the robot's ROS topics by default.

Implementation details are shared during evaluation.

One interaction model, two surfaces

Human-computer interaction on a screen and human-robot interaction on a robot run on the same structured record.

On a screen

What runs on a screen today, with the status of each part
PartWhat it doesStatus
AI roleplay practiceAn AI plays the other person, in character, by voice. A written breakdown when the session ends.Live
Drills20 short formats built from the same material, including a phishing test that arrives by real email.Live
Team viewEach person's practice activity and session scores over time. Never the conversations.Live
Partner embedPractice inside your own learning platform, through an iframe and a signed sign-on link, with a summary of each completed session, including strengths and areas to improve, posted back by webhook.Available on request
Briefings (Insights)A weekly briefing on what changed in your documents, every quote checked against its source, approved by a person before it is sent.In development
Staff answersAnswers grounded in the documents you upload, in the browser.In development

On a robot

  • EMORI practice on a Reachy MiniIn pilot
  • ROS4HRI topics on the robotIn pilot
  • Policy answers that quote your documents or declineIn development
  • NGSI-LD session export for FIWARE analyticsIn development

Built on open conventions

On EMORI, interaction data is designed to be published on the robot as ROS4HRI person topics, with identities replaced by codes before anything is stored.

  1. Mohamed, Y. and Lemaignan, S. ROS for Human-Robot Interaction. IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Prague, 2021, pages 3020 to 3027.

  2. REP-155, the ROS convention for human perception in robotics, based on ROS4HRI. Status: draft.

  3. ARISE, an EU co-funded program for human-robot interaction on ROS 2 and FIWARE. EMORI was selected in its second open call.

  4. Five years of doctoral research in social AI and human-robot interaction at KTH Royal Institute of Technology, by PokaMind's co-founder, Dr. Youssef Mohamed.

The questions your security team will ask

  • The products described on this site are B2B only, for working adults 18 and over.
  • Security documentation, sub-processors and data processing terms come with the evaluation.

Questions that come up in review

Does it profile our employees?
It scores each person's communication in practice sessions, and managers can see those scores over time. That is automated processing of personal data, so it belongs in your DPIA, and we provide what you need for it during evaluation. It is designed not to infer emotions or personality, it does not identify anyone from their face or voice, and it is not built for hiring, promotion or performance decisions.
Where does the data live?
The platform is hosted in the EU on Google Cloud, encrypted at rest and in transit, and handled in line with GDPR. Speech transcription and language analysis run through AI sub-processors that may process data outside the EU; they and the transfer safeguards are listed in the data processing terms we provide during evaluation.
Who sees what?
The person who practiced sees their full breakdown. In PokaMind, managers see each person's practice activity and session scores over time, never the conversation, the transcript or the breakdown. When PokaMind is embedded in your own learning platform, a session summary is sent there and your access rules apply.
Do we have to move to a new system?
No. People use PokaMind in the browser, and partners can embed it in their own learning platform through an iframe and a signed sign-on link.
Is it certified?
Our information security program is aligned with ISO/IEC 27001 and maintained internally. It is not a third-party certification.
Where does it stand on the EU AI Act?
It is designed to stay outside emotion recognition: the practice analysis records observable behavior and is not built to conclude how anyone feels. It is built for practice and development, not to rate, rank or screen people. Because managers can see each person's session scores, the AI Act's rules for workplace systems are relevant too, so the intended purpose is stated in writing: practice and development only, never decisions about people.

See the platform on one of your own situations.