Embedded systems · Edge AI · IoT · Linux

Custom intelligent systems. Private by design.

We design custom hardware, embedded software and AI systems for real-world applications—built local-first, with privacy, cost and long-term control considered from the beginning.

01Custom hardware & systems
02Privacy-first architecture
03Local + hybrid AI
04Purpose-tuned LLMs
SYSTEM / EDGE_NODE_01
LIVE · 24.8°C · 3.71V
INPUT
01 Sensor array
IMU · CAM · MIC · ENV
COMPUTE
EDGE MCU 240 MHz
INFERENCE18 ms
POWER142 mW
UPTIME99.98%
OUTPUT
BLE 5.4CONNECTED
MQTTSYNC
LOCAL AIREADY
What we focus on

Purpose-built technology, not generic platforms.

We combine electronics, embedded software, networking, local AI and cloud services into systems designed around one specific application, environment and privacy requirement.

01

Custom Hardware & Systems

Application-specific devices and complete systems, from PCB and firmware to gateway, backend and user interface.

  • Voice recorders & audio devices
  • Video surveillance & security systems
  • Plant watering & garden automation
  • Asset tracking & geolocation
  • General IoT / AIoT products
02

Privacy-First Design

Privacy is an architectural input, not a feature added after implementation. Sensitive data stays local whenever possible.

  • Minimize external data exposure
  • Local storage & local processing
  • Encrypted communication
  • Private infrastructure options
  • Data-flow design & threat reduction
03

Local-First & Hybrid AI

Run inference on-device or on-premise whenever practical, and use cloud AI selectively where it adds real value.

  • On-device / on-premise inference
  • Local + cloud hybrid architecture
  • Lower recurring AI cost
  • Reduced latency & offline capability
  • Raspberry Pi / edge GPU / MCU AI
04
LLM

LLM Fine-Tuning

Adapt open models for a defined workflow, vocabulary, tool set or application instead of relying on a generic assistant.

  • Domain-specific instruction tuning
  • Intent & tool-calling optimization
  • Structured JSON output
  • Private knowledge integration
  • Quantization & deployment optimization
Example applications

Custom systems for physical-world problems.

Representative product directions rather than rigid packages. Each implementation can be adapted to your hardware, deployment environment, privacy level and budget.

01

Voice & Audio Devices

Recorders, voice interfaces, offline speech systems, event-triggered audio capture and private transcription.

AudioESP32STTLocal AI
02

Video Surveillance

Private cameras, local detection, low-latency streaming, event recording and optional cloud relay only when required.

CameraVisionWebRTCEdge AI
03

Security Systems

Doors, presence, alarms, secure wireless links, local decision logic and privacy-conscious remote notification.

BLESensorsAccessAlerts
04

Garden & Balcony Automation

Soil sensing, irrigation control, weather-aware watering, low-power wireless nodes and simple local dashboards.

SoilValveBLEAutomation
05

Asset Tracking

BLE, motion sensing, local gateways and alert logic for assets that should stay within a defined environment.

BLEIMUGatewayTracking
06

Geolocation Applications

GNSS-enabled tracking, low-power reporting, movement-aware sampling and hybrid local/cloud data paths.

GNSSLTELow powerMaps
07

General IoT & AIoT Systems

Custom sensing, connectivity, control and intelligence for applications that do not fit an off-the-shelf product. We can take responsibility for the complete system instead of only one layer.

HardwareFirmwareLinuxMobileCloudAI
Privacy-first AI architecture
DEVICESense + preprocessAudio · Video · Sensors
LOCAL AIPrivate inferenceLLM · Vision · Speech
PRIVATE SERVERStore + orchestrateLinux · RAG · API
CLOUD / OPTIONALEscalate selectivelyOnly when needed

The design principle: keep sensitive data and routine inference local; use external AI only for tasks where the additional capability justifies the privacy and recurring-cost tradeoff.

Application-specific AI

Make the model fit the job.

A smaller well-adapted model can be more useful, more predictable and much cheaper to run than a large generic model.

Fine-tuning for a defined purpose

We can prepare datasets, fine-tune open LLMs and optimize them around application-specific commands, terminology, workflows and structured outputs.

  • Domain language and specialized vocabulary
  • Intent detection and constrained assistant behavior
  • Reliable function / tool calling
  • JSON-schema compatible outputs
  • Quantized deployment for local hardware
APPLICATION_MODEL / PRIVATEQ4 · LOCAL
user:
"Start watering zone 3 for 8 minutes."

model:
{
  "intent": "START_IRRIGATION",
  "zone": 3,
  "duration_min": 8
}

execution:
✓ schema validated
✓ policy checked
✓ local controller invoked
✓ no cloud request required
How we work

From requirement
to deployed system.

01

Define the constraint

We turn product goals into measurable engineering requirements before choosing technology.

02

Prototype the risky part

We prove power, RF, latency, model performance or architecture early—before polishing the easy parts.

03

Build the system

Hardware, firmware, application and infrastructure evolve together instead of becoming isolated workstreams.

04

Prepare for reality

Logging, recovery, updates, test points and maintainability are designed in rather than added afterwards.

INDEPENDENT ENGINEERING ATHENS · EUROPE

Small team.
Wide technical range.

ViaCat Technology develops custom connected and intelligent systems across electronics, embedded software, Linux, networking and local AI.

Our strongest fit is a project where privacy matters, off-the-shelf products are not enough, or hardware and AI need to be designed as one system.

HW Electronics
FW Embedded
AI On-device
SYS Linux / Cloud
Start a project

Tell us what you want to build.

Share the application, constraints and privacy requirements. Your message is stored in our Cloudflare D1 database and forwarded through our private webhook workflow.

01 Useful details

What the system should do, where it will run, expected volume, power/connectivity limits and any privacy requirements.

02 Privacy by default

We do not need sensitive data in the first message. Please describe the problem rather than sending confidential datasets or credentials.

03 Human response

The form uses Cloudflare Turnstile to reduce automated spam while avoiding traditional image CAPTCHAs.

By submitting, you agree that we may use the information to respond to your enquiry. Do not include passwords, API keys or highly sensitive personal information.