Specialist applications, AI integration, simulation – for every field in which knowledge has to be cast into software
running software, not slides: you see every intermediate state running.
Your dream projects built in weeks instead of years.
One domain expert plus agentic AI – from the task to the production-ready application.
Who is behind it: Dr. Herbert Schuster, domain expert and AI orchestrator – three decades as an independent software maker and project manager (Schuster&Hwesta IT GesmbH, insurance and medicine), specialist in diagnostic and interventional radiology. One contact instead of changing staff – and a clear no when a project does not hold.Write directly
Topics of the publications: interventional radiology, ultrasound and CEUS, prostate MRI and fusion biopsy, tumour ablation.
You name the task from your field. I build it with agentic AI as the tool, review every result myself and hand over running software – including tests, help, installer and documentation. You decide, the AI builds: in one loop of requirement, implementation, test and review on screen – at the pace of vibe coding, Scrum and Extreme Programming (XP), secured by domain expertise, tests and sign-off.
Agentic AI plans, implements, tests and corrects on its own – in loops led by a human. Every result is a verifiable proposal; decisions that matter are made by a human. Effort scales with the number of domain decisions, not with lines of code.
You bring domain knowledge, priorities and acceptance. The agentic AI builds, tests and documents. I lead the AI and set architecture and quality rules – deciding is expert work, building is machine work.
Requirement, implementation, test and on-screen review in one loop – in minutes to hours instead of sprints. Discarded routes are recorded so they do not return.
Self-tests with every change, checks on the running program, recalculation of stored cases: what was right before must stay right.
Application, tests, bilingual help, installer without system entries, reports, licence texts and website – in the same process, in step with the program.
Instead of specifications and bidding rounds: the hardest part of your task, running on your data – at a fixed price agreed upfront. Afterwards both sides know whether and how the project holds.
Task, data, systems and goal – from this the sub-task that decides feasibility.
Exactly this sub-task is built – no mock-up, but a program running on your sample data.
Working build, effort estimate, risk list and a clear recommendation – even if it is: not like this.
For your own project the complete source code with documentation is included in the price – at no extra cost, to develop further, audit or hand over to your own team.
Typical tasks I solve the hybrid way – in any field where expertise has to become software.
Windows desktop, offline, bilingual – with tooltips, shortcuts and help.
Embedding models, GPU/CPU operation, fine-tuning on your data, plausibility checks.
Domain formats such as DICOM, registration, map computation, rule-based assessment with reasons.
Interactive 3D in the browser, planning tools, animated reports.
Word reports with images and reasons, logs, analyses – generated from the results.
Installer without system entries, diagnostic logs, recalculation of old cases, fixes within hours.
Hybrid development speeds up building. Everything that is not building stays as it is.
Conformity, validation and approval take as long as ever – often the larger part of the total time.
Every result is a proposal; the software must make expert review easy, not replace it.
Maintenance, on-call, security updates and training are agreed separately.
The bottleneck is often access to data and systems – we plan for it upfront.
With very large codebases and many parallel teams, coordination dominates.
Regulated data needs a framework: processing agreement, anonymised test data, deletion rules.
If one of these limits affects your project, I will say so in the first conversation.
A domain expert together with agentic AI: the AI plans, implements, tests and documents autonomously in loops; domain experts and I guide it, decide, review on real cases and sign off. The difference to “AI generates code”: toolchain, quality rules, tests and logs are part of the process – which is why maintainable software emerges, not just a prototype.
Anything that translates domain knowledge into software: analysis tools, image and data processing, AI integration, simulations, reports, internal tools, automation. Medicine was only the occasion of the pilot project – see Any field.
First working results within days, mature specialist applications within weeks – depending on the number of domain decisions, not the size of the code. The pilot was built this way in two weeks; what made that possible was the combination of domain and software-engineering knowledge on the human side.
Automatic tests with every change, checks on the running program, a re-run of stored cases, a machine comparison of the program variants and a changelog that also records rejected approaches. Domain acceptance stays with the human.
For your own project the complete source code with documentation is included in the price – as a goody, at no extra cost. Usage rights are agreed per project; third-party components stay under their licences, which I list completely.
Data-frugally: applications run locally, without forced cloud and without telemetry, test data are synthetic or anonymised. Which tools may work with which data is agreed in advance.
With a conversation about task, field and time frame – via the contact form. Then the proof of feasibility in three working days: a first running slice that shows direction and pace.
In loops of hours to days: refine the requirement in dialogue, implement, test, review on a real case, sign off. You see every intermediate state running, not just the final result.
In pace, yes; in care, no. Vibe coding (a term coined by Andrej Karpathy in 2025) means describing software to an AI in conversation and accepting the result without checking the code in detail. I also work in dialogue with the AI – but every result is reviewed, tested and signed off on a real case by the domain expert, with documentation and clearly stated limits.
A concrete offer follows the first conversation – per work package at a fixed price or by effort. The entry point is always the same: the proof of feasibility in three working days at a fixed price.
Yes. Agentic AI reads and understands existing code bases; typical tasks are analysis, adding tests and documentation, targeted extensions or a step-by-step replacement – without downtime.
No. You bring the domain knowledge, I steer the AI. If you want your own team to work the hybrid way, this can be built up alongside – see the skill matrix in three stages.
No and yes. The work is done with agentic coding assistants (Claude Code in the pilot) and Python; the result is ordinary, documented source code with tests, plus logs, help and licence texts. Your team can maintain it with the same tools – or I continue to.
On its own page pilot project Prostate Commander: workflow, features, lesion analysis, fusion biopsy simulation, technology, licence and the demo questions. The simulation runs directly in the browser.
You have a task for which a working specialist solution is needed quickly? Briefly describe field, goal and time frame – I will get back to you with a proposal for the proof of feasibility. The message is delivered directly via the web server of this site, without a third-party form service; if sending via the server is unavailable, your mail program opens with the prepared message.
Demo Prostate Commander: freely usable to try (30 evaluations per installation) – directly via the request form.
deepnet.at