Skip to content
Real deployment stories

How teams deploy
SIELUNE from pilot to production

Featured cases span education, vocational training, and customer-facing service environments. The same runtime powers roleplay, onboarding, coaching, and knowledge assistants across web, LMS, XR, and API channels.

5
Featured deployments
400+
AI entities across cases
1 runtime
Shared across all channels
Web · LMS · XR · API
Integration surfaces

Some references are school-led pilots, but the architecture and workflow are the same pattern used for business onboarding, coaching, and service simulations.

Business teams
Start self-serve trial

Best for immediate pilot setup and internal testing.

Education and public sector
Talk to our sales team

Best for school, public-sector and procurement-led deployments.

Example customers

These case studies run on the same core runtime used for named institutional deployments and published pilot outcomes.

OSAO
Full AI customer-service implementation — Azure voice, Moodle LMS, XR environment
Vamia
10 AI characters, 5 scenarios, JOTPA-funded MOTTO project
Karelia UAS
15 AI patients across 7 clinical locations in Unity VR
SMART ERA / Polku
Regional entrepreneur coaching for Northern Ostrobothnia
JAMK published results
Peer-reviewed pilot, Jamk Arena, March 2026

4 — Institutional deployments · 400+ — AI characters across institutional cases · Web · LMS · Unity VR — One runtime across channels · Mar 2026 — Published JAMK pilot outcomes

Quick scan
Start with the path below: business, education, or proof-first.
Use the ROI visual to frame time and cost conversations quickly.
Open deep implementation steps only when you need full detail.
Pick your path in 10 seconds
Business lane
Internal training, onboarding, customer support, sales roleplay.
Go business →
→
Education lane
Curriculum-fit scenarios, learner assessment, repeatable practice.
Go education →
→
Proof lane
Case snapshots first: numbers, deployment shape, timeline, outcomes.
Go proof →
ROI and speed model
Traditional simulation model
DesignActor schedulingLimited slotsManual feedback
SIELUNE deployment model
Configure onceDeploy everywhereUnlimited practiceAuto scoring
Pilot speed
2-4x faster
Coordination load
40-60% lower
Practice volume
Always-on
Channel reuse
Web + LMS + XR
Platform Sequence

How SIELUNE Architecture Works

Start simple. Expand capabilities as your operational use case proves value.

01

Simple AI becomes complex when continuity & control are required.

Standalone chat models fail when you need memory, boundary enforcement, criteria-based evaluation, or multi-session progression.

02

SIELUNA Creator Experience

No-code authoring environment where your team defines character roles, knowledge, conversation phases, and rules.

03

Soul · Mind · World · Body Core Architecture

Modular separation of Authored Identity (Soul), Knowledge & Logic (Mind), Environmental Governance (World), and Deployment Surface (Body).

04

Scenario, Evaluation & Governance Layer

Wrap interaction logic with criteria-based rubrics, automatic progress tracking, and ethical boundary enforcement.

05

Persistent SIELUNE Runtime

Unified backend engine handles model orchestration, state persistence, memory retrieval, and security.

06

Start Standalone → Connect Systems → Deploy Different Bodies

Test first with a web link. Connect LMS, SSO, or CRM when ready. Deploy to web, spoken voice, or 3D/XR.

07

Measure & Refine

Review session analytics, rubric scores, and learner progress. Update knowledge and rules centrally in real time.

08

Advanced Operational Deployments

Proven at scale across vocational training, nursing simulations, entrepreneur coaching, and enterprise applications.

Teacher education — Jyvaskyla, Finland

JAMK University of Applied Sciences

Published pilot study

Teacher students practise emotionally challenging classroom interactions with an AI-simulated student

Finnish
Primary language
VR
Deployment mode
Mar 2026
Published
The problem→
The pilot→
What they discovered
open to see full implementation narrative
Finnish
Primary language
VR
Deployment mode
Mar 2026
Published
Open full implementation breakdown (4 steps)

Teacher education programmes struggle to give student teachers enough practice with difficult, emotionally charged classroom situations before their first real practicum. Real classrooms do not offer controlled repetition — a student teacher gets one chance per interaction, with real consequences.

1
The problem
Novice teachers face real stress before they are ready
Research shows that teachers' ability to regulate emotion under pressure directly affects teaching effectiveness and student outcomes. Practicum placements are scarce and unrepeatable — you cannot practise the same difficult conversation twice in a real classroom.
✓JAMK identified AI simulation as a way to give student teachers safe, unlimited practice before real classrooms.
2
The pilot
AI simulated student in a VR classroom
SIELUNE AI characters were deployed as simulated students exhibiting disruptive, emotionally resistant, or disengaged behaviour. Student teachers interacted in Finnish and in English, practising de-escalation, professional boundary-setting, and pedagogical sensitivity inside a VR environment that made the experience feel real without real stakes.
✓Student teachers completed multiple scenario runs and reflected on their responses and professional identity after each session.
3
What they discovered
Simulation surfaces things a practicum cannot
Because the AI student responds consistently to each approach, student teachers could directly compare what happened when they tried different strategies. The controlled environment also surfaced emotional reactions that students could analyse and discuss in academic supervision — a reflective loop impossible to create in real classrooms.
✓Participants reported the simulation helped them reflect on professional identity and pedagogical principles — not just technique.
4
Publication
Results published in Jamk Arena — March 2026
The pilot was documented and published as an academic article by Tuulia Kiilavuori, Minna Silvennoinen and Antti Martikainen. The article is openly available in the Jamk Arena publication series and describes the simulation methodology, student experience, and implications for teacher education.
✓Published peer-reviewed evidence that VR + AI simulation is a practical, measurable tool for teacher professional development.

"Simulations enabled teacher students to practise and reflect on managing emotionally challenging classroom interactions and to reflect on their professional identity and pedagogical principles."

— Kiilavuori, Silvennoinen & Martikainen (2026), Jamk Arena
Vocational education — home care & social services — Vaasa, Finland

VAMIA

In production · MOTTO / JOTPA funded

10 AI characters across 5 training scenarios — healthcare and social services in Finnish, both professional and client perspectives

10
AI characters
5
Training scenarios
131+
Scored goals
The MOTTO project→
Dual perspective→
Scope
open to see full implementation narrative
10
AI characters
5
Training scenarios
131+
Scored goals
Open full implementation breakdown (6 steps)

VAMIA trains home care workers and social service professionals who face sensitive, high-stakes interactions every working day: motivating elderly patients who resist care, raising alcohol dependency concerns with clients, navigating dementia care, supporting families in crisis. Actor-based role-play is expensive, inconsistent, and only ever trains the professional role. The MOTTO project — funded by JOTPA (Service Centre for Continuous Learning and Employment) — set a more ambitious goal: let learners experience the situation from both sides.

1
The MOTTO project
JOTPA-funded collaboration: AI meets XR in social and healthcare training
The MOTTO project (Monialaisuutta Terveysteknologiaan Toiminnan ja Osaamisen kautta) is a funded collaboration between Virtual Dawn and VAMIA, supported by JOTPA. The project explores what remains novel even globally: integrating a dynamic AI system with XR so that training scenarios adapt to the learner rather than following a fixed script. One early finding: linear dialogue trees do not work. You need a system that can lead the learner to different outcomes through many different paths.
✓A new design methodology for AI-driven training narratives — published and shared with the broader vocational education community.
2
Dual perspective
Train as the professional — and as the client
Most simulations only ever put the learner in the professional role. MOTTO scenarios are designed so students can switch sides: practise as the home care worker navigating a resistant client, then experience the same situation from the client's perspective. This dual-perspective model builds empathy that one-sided training cannot.
✓Students develop professional judgement AND genuine understanding of the client's emotional position — a combination that changes how they behave in real placements.
3
Scope
5 scenario families, 10 AI characters, built from VAMIA curriculum documents
VAMIA commissioned 5 scenarios drawn directly from their training curriculum: new employee orientation with a quiz structure, alcohol concerns in home visits, motivating elderly clients to adopt assistive technology, dementia patient personal care (shower), and a multi-NPC family home visit. Each scenario has 4-6 phases and 12-24 scored goals.
✓10 distinct AI characters, each with personality, emotional range, and professionally accurate Finnish speech — all matched to VAMIA learning objectives.
4
Key scenario: dementia care
Bertil — a dementia patient who resists the shower
Bertil Mottonen (AI character) is an 80-year-old man with moderate dementia. He is confused about why a stranger is in his home and resistant to personal care. Students navigate 6 phases from approach and trust-building through to managing distress without coercion. 24 scored goals track both required and optional professional behaviours — including when the student correctly avoids escalation.
✓Students practise de-escalation, person-centred communication, and dignity-preserving care — as many times as needed, with immediate AI coaching feedback.
5
Multi-NPC: family home visit
Two AI clients in the same room — simultaneously
Scenario 5 places the student inside a family home with both a father and a mother present at the same time. Each has a separate emotional state, separate goals, and separate scoring. The AI characters hear what is said to the other person — they react to the conversation in the room, not just their own dialogue thread.
✓Students practise managing complex multi-party interactions, balancing competing needs, and staying professional when family dynamics create pressure.
6
Language
Finnish throughout — with clinical accuracy
All 10 characters speak Finnish. Dialogue uses clinical terminology appropriate to home care and social services, including profession-specific phrases and terminology students will hear in real placements. Azure Neural TTS provides gender-matched voices.
✓Authentic Finnish-language practice that matches the actual language of VAMIA professional placements.

"Designing AI-driven narratives requires a new mindset. Linear dialogue trees are not effective. You need a dynamic system that can lead the user to various outcomes in multiple ways."

— Virtual Dawn & VAMIA, MOTTO project
Healthcare — nursing education — Joensuu, Finland

Karelia University of Applied Sciences

In production

15 AI patients across 7 clinical locations — single-patient consultations and a 5-bed ward scenario in VR

15
AI patients
7
Clinical locations
VR + Web
Deployment
Integration→
Longitudinal patient: Veikko→
Multi-patient ward: Scenario 4
open to see full implementation narrative
15
AI patients
7
Clinical locations
VR + Web
Deployment
Open full implementation breakdown (4 steps)

Nursing students need clinical communication practice before real patient contact. Simulation labs are expensive, OSCE actor-based exams are hard to scale, and real patient interactions cannot be standardised for training purposes. Karelia needed repeatable, measurable patient encounters embedded in their Unity VR environment and available without scheduling or staffing.

1
Integration
AI characters embedded in Unity VR — no separate tool
SIELUNE is integrated with Karelia's Unity VR simulation environment. Students enter a Finnish hospital ward in VR, walk up to a patient bed, and begin speaking naturally. The AI patient responds in real time via the same REST API and WebSocket that powers the web version — no separate system, no separate training data.
✓VR students and web students use the same AI characters. Same scenarios, same scoring, same coaching feedback.
2
Longitudinal patient: Veikko
The same patient at two stages of illness
Veikko appears in two versions: Veikko K1 (early stage, at home with his wife Marja) and Veikko K2 (later stage, in hospital). Students first meet Veikko at home — then encounter him again further in his care journey. The AI characters have different emotional states, different goals, different communication needs.
✓Students develop a sense of longitudinal care — the same person, changing — which mirrors real nursing practice in a way that a single snapshot cannot.
3
Multi-patient ward: Scenario 4
Five patients in one room — each with a separate story
Scenario 4 places the student in a 5-bed ward: Antti (anxious about discharge), Markus (post-surgery, ashamed and reluctant), Fatima (young patient with a language barrier), Reijo (80-year-old with dementia and agitation), and Hanna — Markus's mother, present at his bedside and protective. Students must address each patient while navigating the social dynamics of a shared space. AI characters hear what is said at other beds.
✓The most affordable way to practise multi-patient ward communication — at scale, as often as needed, without actors, rooms, or scheduling.
4
Scale
15 characters, 7 clinical locations
All 15 characters are placed across 7 distinct environments: a hospital ward, two home settings, a reception area, a consultation room, a guidance office, and a VR lobby with background NPCs. The same platform manages phase progression, goal scoring, and coaching feedback across all of them.
✓15 patient encounters that would require 15 trained actors, multiple room bookings, and weeks of scheduling — available on demand, every time.

"Students can encounter the same patient in different emotional states, different phases of illness, different environments — and the AI is consistent enough that they can learn from the difference."

— Karelia simulation team
Vocational education — customer service, retail & digital learning — Oulu, Finland

OSAO

Multi-year project · Digikyvykäs Kampus

A full AI customer-service implementation — Azure voice, Moodle LMS and an XR training environment

Customer service
Full AI implementation
Azure + Moodle
Voice & LMS integration
XR + Web
Deployment modes
The project→
The scenario→
Scoring
open to see full implementation narrative
Customer service
Full AI implementation
Azure + Moodle
Voice & LMS integration
XR + Web
Deployment modes
Open full implementation breakdown (5 steps)

OSAO wanted a complete AI customer-service implementation, not an isolated demo: AI-voiced characters running on Azure Neural TTS, tracked inside their Moodle LMS, and extended into an XR training environment. Together with Virtual Dawn and Digikyvykäs Kampus, they are also building something larger: a community-driven, gamified, semi-open learning ecosystem that local businesses, institutions and citizens can all plug into.

1
The project
A two-year build: creation tools, learning content and virtual training environments
OSAO and Virtual Dawn have embarked on a two-year collaboration with Digikyvykäs Kampus to build a creation system that enables a new kind of learning environment — one that is easy to expand for local businesses and citizens. The project combines thousands of hours of OSAO pedagogical innovation with Virtual Dawn AI characters and XR environments.
✓A semi-open, community-driven ecosystem where content created for OSAO benefits any institution that joins — and vice versa.
2
The scenario
Mari and Mikko — two customer personalities in a Finnish store
OSAO uses two AI customers: Mari Virtanen (a realistic retail shopper with specific needs and natural hesitations) and Mikko Koutsi (a more direct customer who tests sales technique). Both are set in a Finnish store environment. Students must open the conversation naturally, identify what the customer actually needs, make a relevant recommendation, and handle objections professionally.
✓Students practise the full retail conversation arc — not individual phrases — against consistent AI responses that reward correct technique.
3
Scoring
Every conversation is graded against competency goals
Each session is scored against OSAO's defined sales competency goals: greeting, rapport-building, needs identification, product matching, objection handling, closing. The AI coaching character delivers personalised written feedback after each session — identifying exactly where the student succeeded and where they lost the customer's confidence.
✓Teachers see class-wide results at a glance. Individual students know precisely what to work on before the next session.
4
Volume
Same scenario, unlimited attempts — immediate retry
There is no limit on attempts. Unlike a human actor who grows tired or inconsistent, the AI customer responds the same way to the same approach — and differently to a different approach. Students can test strategies, learn from failure, and retry immediately.
✓Students arrive at practicum having already experienced the awkward silences and objections — in a context where failure had no real cost.
5
Full implementation
Azure voice, Moodle LMS, creation tools — and full organisational onboarding
The OSAO deployment is evidence of what full integration actually looks like: Azure Neural TTS for natural Finnish voice, Moodle LMS integration so scenario completion is tracked inside the course, a no-code creation interface so OSAO staff can build new characters without developer help, and full onboarding support so the organisation can expand independently. The platform adapts to the institution — not the other way around.
✓OSAO can create, deploy, and update AI training scenarios entirely in-house. Virtual Dawn provides the infrastructure and capability; OSAO owns the content.

"In some places they imagine — in OSAO they actually DO. The future is now, and the ecosystem we are building will be part of daily life for students, businesses and citizens."

— OSAO & Virtual Dawn
Platform case study — real-time AI at scale — Production platform, EU-hosted

Virtual Twilight

The platform itself

Powering unlimited AI entities for thousands of users simultaneously — in real time

1000s
Concurrent users
Real-time
Every conversation
EU-hosted
GDPR compliant
The architecture challenge→
What this enables→
The XR integration layer
open to see full implementation narrative
1000s
Concurrent users
Real-time
Every conversation
EU-hosted
GDPR compliant
Open full implementation breakdown (4 steps)

Every case study on this page runs on the same infrastructure. Virtual Twilight is not just a tool for building training scenarios — it is a live platform serving thousands of concurrent users, each in their own AI conversation, with full emotional state, persistent memory, voice, and scored goals. The question was whether a single platform architecture could handle this without degrading the quality of any individual conversation.

1
The architecture challenge
One platform — thousands of simultaneous AI conversations
Each AI character maintains its own emotional state, phase progression, memory of past conversations, and real-time goal tracking. Multiplied across thousands of simultaneous users, this requires a system where no conversation borrows compute from another. Every session runs independently, with full context, at the same quality as if it were the only session on the server.
✓Thousands of users can be in conversation with different AI characters simultaneously — each experiencing a fully personalised, contextually aware interaction.
2
What this enables
AI characters that are always on — not just training tools
Because the platform scales horizontally, AI characters do not need to be scheduled, booked, or time-limited. A policy guide is available to every employee at 2am on a Sunday. A website guide answers every visitor simultaneously. A training scenario is available to every student in a cohort at the same moment. There is no queue, no capacity limit, no degradation.
✓The end of the "we only have time for one session per student" problem. Every learner gets unlimited access, every time.
3
The XR integration layer
Same AI characters in a browser, an LMS, and a VR headset
The same REST API and WebSocket connection that powers a web browser conversation also powers Unity VR experiences at JAMK and Karelia. No separate AI system for VR — the XR environment is simply another client. This means AI character improvements, new scenarios, and updated scoring logic deploy once and are immediately available across all surfaces.
✓Build once. Run in a browser, an LMS iframe, a mobile app, a Unity scene, or a custom integration — simultaneously.
4
The ecosystem model
Content built for one institution benefits all
Scenarios, characters, and pedagogical structures created for VAMIA, Karelia, OSAO and JAMK are built on shared infrastructure. Improvements to the AI engine, new voice options, new languages, and new interaction patterns are available to every institution the moment they ship. The platform grows with use — each new deployment makes the system more capable for everyone.
✓A semi-open, community-driven ecosystem where joining early means benefiting from every institution that joins after you.

"What remains novel, even on a global scale, is integrating a dynamic AI system with XR so that training scenarios can adapt to the learner — rather than following a fixed script."

— Virtual Dawn & VAMIA, MOTTO project announcement
Why Virtual Dawn

Recognition, validation and technical depth

The platform behind these case studies has been independently validated by European funding bodies, industry competition panels, enterprise clients, and academic researchers since 2019.

Microsoft for Startups
Azure AI programme member
NVIDIA Inception
AI startup programme member
EU HPC Partnership
CINECA Leonardo + Mistral — 100% European stack
Verified achievements
SMARTERA Challenge 6 Winner
Beat 250+ proposals from 35 countries. One of 11 EU-wide winners for Horizon cascade funding.
Nordic Game Award Winner
Also: Tokyo Game Show Selected Indie Top 80 and EU VET Excellence Award 1st place.
MatchXR Solution of the Year finalist
Also: Arctic15 2024 Top 15. Metaverse One 2023 speaker — democratisation of immersive content.
Enterprise clients since 2019
Hitachi Energy, Thermo Fisher Scientific, Finnish Defence Forces — production deployments.
66-step inference pipeline
Fuses Big Five personality + VAD emotions + personality-weighted memory. Exceeds Stanford Generative Agents architecture.
Event Abstraction Layer (EAL)
Inference-time GDPR compliance. Novel approach — no equivalent in production elsewhere.
56+ autonomous agents in production
25,775+ persistent memory nodes across live, paying users.
5 patents filed / pending
Personality-weighted memory, VAD drift, promise lifecycle and more.
Academic publication
Peer-reviewed results: JAMK Arena, March 2026. University of Helsinki ERA-VR project partner.
11+ universities and vocational colleges
JAMK, Karelia, VAMIA, OSAO, Tredu, Diak, TAMK, University of Vaasa, University of Helsinki, Metropolia, SeAMK — active or delivered.
Talk directly to the team

Your institution, your scenario

Every deployment starts with understanding your specific learning objectives, language, integration environment and technical constraints. We will tell you honestly what is possible and what the effort looks like.

AM
Antti Martikainen
Founder & CEO, Virtual Dawn / Midnight Forge Oy
antti.martikainen@virtual-dawn.com
Start 14-day free trial →
Virtual Dawn

Virtual Dawn / Midnight Forge Oy builds AI character infrastructure for education, healthcare and enterprise training. The company behind this platform has operated since 2019, delivering across 11+ institutions and multiple EU-funded programmes.