Edge AI in production — not just in the lab
Physical AI that works at the edge.
PRAESC builds adaptive, uncertainty-aware AI systems that run on embedded hardware. No cloud dependency. No vendor lock-in. Your data never leaves your facility.
Built on a decade of EU-funded industrial R&D
Industry verticals
We deploy production-grade Physical AI across four domains.
Robotics
Autonomous manipulation, fully offline.
- ▲ All inference on-device — no cloud calls, no IP leakage
- ▲ Confidence-gated execution — human in the loop when uncertainty is high
- ▲ Validated at industrial process ecosystem
Automotive
Adaptive control that learns in real time.
- ▲ Sub-5 ms step latency on Jetson-class hardware
- ▲ Continuous adaptation — zero retraining required
- ▲ Validated on Tier-1 OEM test vehicles
Smart Farming
On-premise AI for precision cultivation.
- ▲ Predictive models for yield and irrigation control
- ▲ Single compute box per facility — no cloud dependency
- ▲ Active advisory engagements with commercial growers
Manufacturing
Quality inspection and predictive maintenance.
- ▲ False-positive rate under 3% at line speed
- ▲ 48–72 h early warning on equipment failures
- ▲ Models 3× smaller than baseline — runs on your existing hardware
How we work
Three principles underpin every system we build.
Edge-Native
Inference runs where data is generated — on-device, offline-capable, with no cloud latency, connectivity constraint, or IP leakage.
First Principles + Physical AI
Bayesian reasoning and principled world models built into every system — structured, explainable, and calibrated. Systems that quantify what they know and flag what they don't.
Trustworthy by Design
Uncertainty quantified, human-in-the-loop by default, EU AI Act aligned. Systems that know what they don't know.
Selected work
Real deployments, real constraints. Results validated in simulators and on-site at industrial testbeds.
Autonomous Manipulation for Process Plants
Physical AI system combining a compact Vision-Language-Action model with a physics-grounded Bayesian confidence loop for autonomous manipulation operations. Runs entirely on Nvidia Jetson-class hardware — no cloud dependency, no data leaves the site.
"The full manipulation pipeline on a single edge device, zero cloud dependency — PRAESC delivered exactly what our industrial partners require in terms of on-premise AI."
— Dr. Dominic Gorecky, Head of Swiss Smart Factory
Real-Time Adaptive Damper Control
Active Inference controller for vehicle suspension, running in real-time on Jetson hardware. The controller continuously adapts to road conditions and driving style without retraining — personalising suspension response in real time.
About PRAESC
PRAESC GmbH is a Swiss specialist firm founded by Dr. Miguel de Prado (PhD, ETH Zürich). We specialise in the end-to-end deployment of Physical AI — from algorithm design and edge-hardware optimisation to industrial validation.
Our edge: a rare combination of embedded-systems depth, Physical AI know-how grounded in a decade of delivering R&D projects across robotics, automotive, and manufacturing verticals.
Our vision →- ▲ Founded 2024, Biel, Switzerland
- ▲ Active industrial engagements
- ▲ 5+ EU Horizon projects delivered
How we start
Most engagements follow the same three steps — from first call to running system in under 12 weeks.
Scoping call (1 week)
We map your use case, constraints, and existing data. You get a written technical scope — problem definition, proposed approach, and a realistic delivery timeline — at no cost.
Proof of concept (3-5 weeks)
A working prototype on your data or a representative dataset. Fixed scope, fixed price. You own everything produced — code, models, documentation.
Production-grade deployment (4–6 weeks)
Hardened, tested system deployed on your hardware. Handover includes full documentation, a trained model, and a 90-day support window.
Is PRAESC the right fit?
We work well with
- ✓ Teams with an engineering lead who owns the hardware integration
- ✓ Projects where the AI layer is the bottleneck, not the infrastructure
- ✓ Organisations needing specialist depth alongside internal capability
Not the right fit if you need
- ✗ A large team running multi-year maintenance contracts
- ✗ Out-of-the-box SaaS — our work is bespoke and project-scoped
Common questions
Do I need my own hardware? +
What does a fixed-price PoC actually include? +
Can you work under NDA? +
Do you outsource? +
What happens to our data? +
Have a challenge in mind?
Tell us what you're building. We'll tell you how to get it to the edge.
Let's talk