Skip to content
Contact

Custom software · Simulation · Systems in Rust

Applied research

AI automation for businesses. Less routine. More time.

elsyQ automates time-consuming business processes with artificial intelligence and custom software. We connect your existing systems and build solutions that reduce manual work. Software that computes, simulates and acts.

Discuss your project What we build

What elsyQ does

Show us where your company is losing time.

elsyQ builds the right solution from AI automation, intelligent agents and custom software. For demanding tasks we rely on numerical methods, simulation and machine learning in Rust, C and C++. And we bring the solution to where you work: into your web interface, into your API or directly onto your machine.

  1. From the problem to the right tool.

    Tell us which problem you want to solve. Then we decide what to solve it with: an algorithm, a simulation, an AI agent or a combination of them. The technology follows the task, never the other way round.

  2. From fixed rules to AI automation.

    Every automation starts with determinism. Whatever can be solved with fixed rules and algorithms we solve that way, because the result can then be checked and comes out the same every time. AI and agents only come in where fixed rules reach their limits.

  3. From the formula to the simulator.

    We build simulators and compute cores ourselves, for electronic circuits or physical processes, for example. Written in Rust, fast enough for large models and with results whose accuracy we can state. Our own SPICE simulator, Circuit Forge, is built exactly this way.

  4. From the web interface to the machine.

    A result is only useful once someone can work with it. That is why we build the whole chain: web interface and API in Laravel, connection over MQTT, firmware on the machine. All from one team.

What we build

For electronics and semiconductor companies, for mechanical and plant engineering, for energy and measurement technology. And just as much for any other company, whatever the industry. If you have a process that eats too much time, you are in the right place with us.

agent harness

An agent harness for your business

A language model on its own gets no work done. It needs a frame: tools to operate your systems, permissions that define what it may do, and approvals before anything important happens. We build that frame on your infrastructure, shaped around your processes.

Every step an agent takes is logged. Tests show whether it does its job right before it takes that job over in operation. And the model behind it stays replaceable, local or through an API.

automation

AI automation

Checking quotations, reading datasheets, reconciling orders, producing reports. If a process can be described clearly and still costs hours, it can usually be automated. We build it as agents with access to your systems, either on local models or through OpenAI, Anthropic and Mistral.

AI automation in detail
numerics

Numerics & mathematical solvers

Linear algebra, sparse systems, integration of ODEs (ordinary differential equations) and DAEs (differential-algebraic equations), Newton methods, FEM (finite element method), signal processing, optimisation. Scientific computing where somebody can state how large the error is.

simulation

Circuit simulation & SPICE

Solver cores for transient, AC and DC analysis. Plus device models, convergence strategies and netlist tooling. Still working when the circuit refuses to cooperate.

ml

Machine learning on technical data

Anomalies in measurement series, classification of signals, predictive maintenance. Plus surrogate models that make an expensive simulation orders of magnitude faster. Trained on your data, judged by your metrics.

systems

High-performance & embedded software

Rust, C and C++ from firmware up to the compute core. Real-time behaviour, memory budgets, SIMD, parallelism. Software that stays predictable at full load.

web

Web ↔ hardware

Laravel applications, APIs and dashboards wired to your equipment through an MQTT broker. With roles, permissions and an audit trail. And an interface that also works on the night shift.

Between the model and the machine: AI, numerics and real systems in one codebase.

elsyQ

Research

applied

We research and develop new methods in machine learning, mostly with neural networks. For our own tools, and as contract research for you.

You have a problem with no ready-made solution yet, or the obvious one is too slow or too imprecise? Then we first read what the state of research offers. We rebuild the method and measure it against your data, next to the procedure you use today. What holds up goes into production. What does not, you hear about from us.

We do not publish papers. We ship methods that hold up in operation.

What we work on

  1. Surrogate models

    A network learns the behaviour of an expensive simulation model and then answers in milliseconds instead of minutes. It is trained on the data of the solver we built ourselves. We also tell you across which range of parameters the prediction holds and from where on you have to compute again.

    fθ(x)≈f(x),x∈Ω

    Ω: the parameter range in which the prediction holds

  2. Physics-informed learning

    A model that already knows the conservation laws needs less data and stays plausible where nobody has ever measured. We write the physics into the loss function as a constraint instead of having the network learn it from examples.

    ℒ=ℒdata+λ·ℒphysics

    λ weights how strictly the physics applies

  3. Hybrid AI: learning and reasoning

    A language model guesses when it runs out of knowledge. That is why we couple it to rules and solvers that check every result before it reaches your system. The checking part stays deterministic, so afterwards you can retrace why a decision came out the way it did.

    ŷ=LLM(x),y=R(ŷ)

    R: rules and solvers that check every answer

  4. Measuring uncertainty

    A number without an error bar is an opinion. We estimate how confident a model is on a concrete input and let it hand over as soon as that confidence drops too low: to an exact solver or to a person.

    σ(x)>τ⇒hand-over

    σ: estimated uncertainty, τ: your threshold

How we work

  1. Analysis

    We look at your process and work out what the solution has to deliver. After that it is clear whether the project is worth it for you.

    → brief

  2. Prototype

    Within a few weeks there is a first working version, tested with your real data or, where it must not leave the building, with realistic test data.

    → running prototype

  3. Presentation

    We show you the prototype and go through the results together. Only once you are convinced do we move on.

    → your sign-off

  4. Production

    We bring the solution into your day-to-day operations, with monitoring and documentation. After go-live we stay on board.

    → live operation

Two principles

If we cannot make it better, we say so.

Sometimes the existing process is already good enough. Sometimes the problem does not belong in software at all. We say that in the first conversation instead of turning it into a project.

Verifiable where it matters.

Where money, safety or a measurement depends on it, no language model makes the call. An algorithm runs there whose result can be reproduced and checked. A model only comes in when the numbers show that it performs better.

Algorithm or language model: when we use which

Our process in detail

Technology

One team across the whole stack. From the compute core and the firmware to the model and the interface.

Languages & systems

across every layer
  • Rust
  • C / C++
  • Python
  • PHP 8.5+
  • Laravel
  • Linux

  1. Web & IoT

    Interface and connection

    • Livewire
    • REST · WebSockets
    • MQTT broker
    • IoT dashboards
    • WebAssembly

  2. AI & machine learning

    Models and agents

    • PyTorch
    • ONNX
    • OpenAI · Anthropic · Mistral
    • Local models
    • Agents · MCP
    • Agent harness
    • RAG · Vector search
    • Anomaly detection
    • Surrogate models

  3. Simulation

    Physics in the computer

    • SPICE
    • Transient · AC · DC
    • Noise analysis
    • Monte Carlo
    • Device models
    • Netlists
    • Physical models

  4. Numerics

    The compute core

    • Linear algebra
    • Sparse solvers
    • ODE / DAE
    • Newton methods
    • FEM
    • DSP · FFT
    • Optimisation

  5. Embedded & microcontrollers

    Firmware and measurement

    • STM32
    • RP2350B
    • ARM Cortex-M
    • Embedded Rust
    • RTOS
    • Firmware
    • Measurement

Frequently asked questions

What companies ask us before the first project.

What exactly does elsyQ do?

elsyQ automates time-consuming business processes with AI and custom software, and builds compute-intensive systems: numerical solvers, simulators such as the SPICE simulator Circuit Forge, and machine-learning models for technical data. For automation the rule is fixed rules first, AI only where rules cannot go further. Compute cores and firmware are written in Rust, web interfaces and APIs in Laravel.

Is elsyQ an AI agency?

No. elsyQ is not a classic AI agency but a software engineering company that does its own research. We work on questions of automation, machine learning and information theory and turn the results into our own methods and tools, such as the SPICE simulator Circuit Forge. We use AI where it demonstrably solves a problem better than a deterministic algorithm. It then sits inside a system with clear interfaces, tests and permissions, not in a chat window in front of an unchanged process.

Does our data have to go to OpenAI or Anthropic?

No. elsyQ can run language models entirely locally, on your own hardware or in your own cloud, so no data leaves the building. Open models such as Llama, Qwen or Gemma are good enough for many automation tasks. Where data protection sets no limit, we also work with OpenAI, Anthropic or Mistral. Protection requirements, latency and cost decide, and the architecture stays built so that switching later remains possible.

Does elsyQ connect its solutions to our existing systems?

Yes, elsyQ connects every solution to the systems you already use instead of placing yet another stand-alone tool next to them. In the office that usually means the ERP, the CRM or the mailbox, connected through REST interfaces. If an older system has no API, we go through its database or its file exports. On the shop floor we talk to machine controls and measuring instruments over MQTT or OPC UA. We settle the interfaces early in the project and test the whole chain at the end.

Which industries does elsyQ work for?

elsyQ works for electronics and semiconductor companies, mechanical and plant engineering, energy and drive technology, and measurement and medical technology. For AI automation the industry does not matter: we automate processes for any company. All that counts is that a process can be described clearly and currently costs too much time.

Why does elsyQ use Rust?

Rust offers memory safety and concurrency without a garbage collector, so the performance of C++ with far fewer classes of bugs. For solver cores, firmware on microcontrollers and long-lived services that is the decisive advantage. Rust is therefore our first choice but not a dogma: web interfaces are built in Laravel, models in PyTorch, analyses in Python, and existing codebases are maintained in C or C++.

What if a process cannot sensibly be automated?

Then elsyQ says so openly before writing a quote. Not every process gets better through software, and some problems are organisational rather than technical. We check in the first conversation whether the effort pays off, and when in doubt we recommend leaving it alone.

Does elsyQ design electronics and hardware?

Yes, elsyQ also develops electronics and hardware when the project is a good fit. Our team has deep knowledge of electronics and hardware-near development: we simulate circuits, write firmware for microcontrollers such as STM32 or RP2350 and build the data acquisition around it. We are strongest when the hardware is part of a system whose software we are responsible for as well. Pure layout services without a software share are not something we offer.

Do we get documentation?

Yes, every elsyQ project is fully documented, because for us documentation is part of the product. It is written alongside the code, not in the last week of the project. Architecture, interfaces, usage and operation are described so that your team can understand the system and develop it further without us. We train the people who will work with it, and the repository and documentation belong to you.

How does a project with elsyQ start?

A project with elsyQ starts with a conversation about your problem. It produces a short brief covering goals, risks and an effort range. Within a few weeks we show you a working prototype, and only after your sign-off do we move into production. We answer enquiries within two working days.

Your question is not here?

Write to us. We reply within two working days.

Ask a question

Let’s talk

A technical problem that needs more than a prompt?

Write us a few sentences about it: domain, constraints, target numbers. You get a first technical assessment within two working days, not a sales deck.

Discuss your project [email protected]