Research project · BR24992975

Digital Twin
of a Food
Processing Plant

IIoT data acquisition, machine learning, schedule optimisation and 3D visualisation — a single platform for real-time management decisions.

Powered by artificial intelligence and Industrial Internet of Things technologies

Digital Egiz Lab Al-Farabi KazNU OpenEgiz · open-source

Technology partner

Siemens — academic programme for KazNU faculties

The full cycle of industrial software: digital design, production-process modelling, engineering analysis, manufacturing management, quality control and simulation of complex physical processes.

DesignSimulation Engineering analysisManufacturing Quality

NX AcademicCore & CAD · CAE & CAM · AM Add-on

CAD/CAM

Simcenter 3DAcademic Bundle

CAE

Simcenter STAR-CCM+Academic Power On Demand

CFD

LMS Test LabAcademic Bundle

Testing

TecnomatixManufacturing Academic Bundle

Manufacturing

Plant SimulationResearch Concurrent

DES

OpcenterAPS · RD&I · EX CR · Execution · Quality

MES/MOM

Platform · open source

OpenEgiz architecture

An open digital-twin stack: devices push data through Eclipse Kanto and Kafka/MQTT into Eclipse Ditto, from where it flows on to storage and visualisation — Grafana, Unity 3D and FMI-standard simulation.

Eclipse DittoEclipse KantoKafka / MQTT InfluxDBMongoDBGrafana Unity 3D · WebGLFMI

Modelling

3D modelling via a Telegram bot

A request in a messenger turns into a finished 3D model of the equipment — no CAD and no special skills needed.

Printing the 3D model — From a Telegram request to a printed 3D model

1

Request to the botThe user describes the object or sends a photo in Telegram.

2

Model generationThe bot builds a three-dimensional model of the equipment.

3

Export and printingThe finished model goes to visualisation or to 3D printing.

${icon_play}Video example Instagram Reel ${icon_doc}Official press release gov.kz

Real-time monitoring

Interactive 3D scene

The laboratory digital twin monitors equipment continuously through a 3D interface. Ovens, mixers and ice machines show their online / offline status right on the models.

${icon_bolt}
Poweractive and reactive power for every unit
${icon_gauge}
Phase currents · cos φpower factor and phase balance in real time
${icon_chart}
Daily consumptionkWh usage for quick visual analysis

Telemetry

Real-time dashboards

Grafana panels for energy use and equipment condition: phase currents, power, load profile and automatic anomaly detection.

Model library

Equipment digital twins

Every piece of equipment — ovens, mixers, ice machines, proofing cabinets — is represented by its own digital twin with parameters, model and status.

Case · energy monitoring

Energy metering on Saiman meters

IIoT energy monitoring system based on Saiman CO-3711 meters

Case · microclimate

Microclimate control: proofing and cooling zones

Microclimate control system — proofing and cooling zones

Case · level sensor

Laser liquid level sensor

Augmented reality

Interactive plant mock-up

The digital twin merged with physical space: point a smartphone at the printed 3D mock-up of the shop floor — or at the real equipment — and see IoT data overlaid on it.

${icon_grad}
Educationa vivid, physically tangible interactive stand for students
${icon_eye}
Productionfast visual diagnostics of equipment right on the shop floor

AR: fault — phase imbalance — Fault: phase imbalance

Data security

IIoT security platform

End-to-end AES-GCM encryption along the whole path — from sensor to gateway. Only verified data reaches storage.

100%

encrypted data

Real-time

integrity validation

AES-GCM

end-to-end encryption

${icon_cpu}
Sensor & client · ESP32data acquisition and end-to-end encryption at the source
${icon_net}
MQTT brokeruntrusted transport — protected against compromise
${icon_shield}
Gateway & validation · Raspberry Piintegrity checks and detection of unauthorised devices
${icon_server}
Storage · InfluxDBonly verified data is persisted

ESP32RPi 4MQTT InfluxDBREST APIAES-GCM

Digital tool for the process engineer

Mobile assistant

A digital tool for the bakery process engineer: orders, doughs, batches and stage tracking — all in one app.

100%

digital record-keeping

−60%

manual labour

Real-time

stage tracking

1C

automatic order import

${icon_kanban}
Process engineer boardorders, forecast, grouping by dough batch
${icon_box}
Batch managementautomatic calculation of trays and loading
${icon_gauge}
Stage trackingKanban with real-time timers
${icon_recipe}
Recipe databasenorms, parameters, integration with orders
${icon_arrow_rs}
1C integrationorders via REST API / OData

Mendix Low-CodeReact Native REST APIOData

Technology

Contactless mouse on Raspberry Pi

Cursor and click control by hand gestures in front of a camera — natively under Labwc / Wayland, without X11.

${icon_camera} Camera ${icon_arrow_r} MediaPipe Hands ${icon_arrow_r} Gesture recognition ${icon_arrow_r} uinput / evdev ${icon_arrow_r} Cursor / Click

MoveOPEN

Open palm — the cursor follows the centre of the hand

ClickPINCH

Pinch of thumb and index finger — a single click

DragFIST

Fist — button held down for drag & drop

MediaPipeuinput / evdevWayland · Labwc 21 hand landmarksSTICK smoothing

Applications

Why a contactless mouse

Controlling a computer where physical contact is impossible or undesirable — three real-world cases.

Medicine

A gloved surgeon controls the software — scans, 3D models — without breaking sterility in the operating room.

Accessibility

Full control of a computer with one-hand gestures — for people with disabilities.

Industry

Operating terminals amid dirt, oil and dust — in work gloves, without removing any gear.

${icon_check}

No contactnothing has to be touched

${icon_check}

Any softwareHID mouse emulation

${icon_check}

AffordableRPi + USB camera

${icon_check}

3 gestureslearned in minutes

Learning platform

lms.digitalegiz.kz

An open-source Chamilo LMS deployment, adapted to the enterprise architecture — staff training and certification on our own infrastructure.

${icon_cloud}
Independenceentirely free of cloud providers
${icon_check}
Automatic knowledge checkstests and certificate generation
${icon_globe}
Available 24/7from any work device

${icon_ext}Open the platform lms.digitalegiz.kz

AI course generator

SCORM Agents

Interactive SCORM courses built from corporate documents — automatically, through a Telegram bot.

PDF / DOCX

Any format

RAG

Retrieval with reranking

Real-time

Telegram bot

SCORM 1.2

Packages for LMS

${icon_doc}
Ingestiondocument parsing, chunking, ChromaDB
${icon_layers}
Architect AgentAI designs the course structure
${icon_spark}
Writer Agentcontent generation with RAG retrieval
${icon_check}
Assessmentautomatic tests derived from the content
${icon_box}
SCORM PackagerHTML + ZIP for Chamilo / Moodle

LangGraphQwen 3.5ChromaDB DoclingSCORM 1.2

Capabilities

Documents → a course in minutes

From a corporate file to a finished interactive course — with no instructional designers and no graphic designers.

${icon_doc}

Any format

PDF, DOCX, PPTX

${icon_layers}

Smart structure

AI proposes the modules

${icon_spark}

Sourced content

RAG retrieval over the documents

${icon_check}

Tests and SCORM

MCQ, True/False → ZIP

3

AI agents
Architect · Writer · Assessment

2

interfaces
CLI + Telegram

RU·KK

multilingual
Russian and Kazakh

100%

automation
from document to ZIP

HR and L&D teams

Fast course creation from internal documents without contractors

Training centres

Scaling course production without instructional designers

Corporate universities

Routine automation in curriculum development

Pipeline · RAG + LLM

Automatic course generation

PDF / DOCX → RAG + LLM → SCORM 1.2

1

Upload

Telegraf.js Telegram bot
PDF / DOCX files
modules, language, model

2

Parsing

Docling (IBM Research)
headings, tables, paragraphs
50+ Cyrillic regexes

3

RAG + Embedding

Parent-Child chunking
qwen3-embed 4B · dim 2560
ChromaDB · Top-6

4

LLM generation

gpt-oss 20B Ollama
gpt-oss 120B vLLM + Ray
Tensor Parallelism · 2 nodes

5

Validation / Export

content deduplication
SCORM 1.2 · PDF · PPTX
Chamilo (REST API)

${icon_cpu} 2× DGX GX10 Spark · 119 GB × 2 · ConnectX-7 400 Gbit/s · Tensor Parallelism across 2 nodes

Deployment diagram

Generating and deploying a course into Chamilo

Telegram bot

Scan it to build a course from your own document.

1

Choose a languageRussian or Kazakh

2

Upload a documentPDF / DOCX and parameters

3

Get a SCORM courseauto-uploaded to Chamilo

Collaboration

OpenTeams

A single open-source platform for team collaboration — a self-hosted replacement for Slack, Drive, Trello and Zoom.

$0

per user

5

services in the stack

12

Docker containers

1

command to launch it

${icon_folder}
Nextcloudfiles, versioning, sharing
${icon_chat}
Rocket.Chatchannels, messages, discussion threads
${icon_kanban}
WekanKanban boards, cards, deadlines
${icon_video}
Jitsi MeetHD video, screen sharing, recording
${icon_lock}
LLDAPunified authentication, SSO

${icon_shield}
Data sovereigntyeverything on your own servers, RU / KZ data residency
${icon_lock}
LDAP single sign-onone account, automatic Let's Encrypt SSL
${icon_server}
Daily backup7-day retention, launched with a single command

${icon_ext}Source code · MIT License github.com/rtzgod/openteamskz

AI for meetings

OpenTeams PRO

AI-powered collaboration — on your own infrastructure.

Metric	Without Pro	With Pro
Meeting transcription	Manual notes	Automatic, real time
Meeting reports	15–30 min of prep	0 min — auto-generated
Searchable archive	~40% retained	~100% — full text
Tasks (Kanban)	Manually	Auto from the meeting context

On-premises deployment

AI on your own servers. Full control over data and models, maximum privacy.

Cloud mode

Our cloud for a fast start. No GPU infrastructure costs, works right away.

Enterprise HPC

Compute cluster

Our own independent infrastructure for high-performance computing and edge device management.

${icon_server}
Compute capacity2× Supermicro · Intel Xeon Gold 6526Y · DDR5 + NVMe · ASUS Ascent GX10 accelerators
${icon_net}
Network configurationMikrotik — traffic management · industrial LAN for IIoT
${icon_cpu}
Capabilitiesdistributed computing across 2 nodes · real-time digital twins · neural network training without cloud latency

Hardware

Key cluster specifications

2×

Supermicro servers

5th

generation Xeon Gold

DDR5

memory + NVMe

GX10

ASUS Ascent accelerators

Hardware platform

Intel Xeon Gold 6526Y — 5th-generation CPU
DDR5 ECC — next-generation memory
NVMe SSD — high-speed storage
ASUS Ascent GX10 — ML/AI accelerators

Network and capabilities

Mikrotik Routing — traffic and security
Industrial LAN — network for IIoT
Edge Computing — computing without the cloud
Digital Twins — twins in real time

Laboratory

Technology infrastructure and R&D

${icon_server}Servers and AI compute

Supermicro SYS-611C-TN4R — Xeon Gold 6526Y, 128 GB RAM, NVMe/SATA SSD
SHIP 103 42U server cabinet
SVC V1500-L-LCD UPS — uninterruptible power

${icon_cpu}Workstations

12 workstations, two of them — AMD Ryzen 9 (16 cores), 192 GB DDR5, RTX 4090
6× ASUS Ascent GX10 — 1 petaflop and 128 GB of memory each for AI workloads

${icon_cube}Robotics and 3D printing

Ackerman Robot — platform with radar and a ROC-OrinNX module
BambuLab X1 Carbon Combo and the large-format ELEGOO OrangeStorm Giga

${icon_eye}VR and visualisation

Meta Quest 3 — mixed and virtual reality
Xiaomi Mi 4K Laser — ultra-short-throw 150″ projector

About us

Digital Egiz Lab

The digital twin laboratory of Al-Farabi KazNU — digital models, machine learning and IIoT for industry.

${icon_globe}Laboratory website digitalegiz.kz ${icon_play}YouTube @DigitalTwinPro ${icon_camera}Instagram @digital_twin_kaznu

Exhibition

InterFood Astana 2025

The laboratory team presented the digital twin of a food processing plant at the sector’s largest trade show — with real-world use cases for digital models, machine learning and IIoT.