About IQAbel
IQAbel works with a small number of clients at a time on production AI projects where the engineering has to be done properly. Medtech, sports, and the smaller organisations that don't have a research lab but still need their AI to actually function.
The studio
IQAbel is an independent AI engineering studio. We design, build, and ship production AI for organisations who need the work to last beyond the demo. Most often that means medtech and healthcare-adjacent companies, sports federations and clubs, and the small-to-mid organisations that don't run a research lab but still need their AI to behave like adult software.
Our engineering practice draws on decades across frontend, product, and platform work — coming up through the era when the web meant making software hold together across browsers that disagreed about almost everything. Current focus is applied AI in production: most recently inside regulated medtech environments and on cognitive architecture work that sits between research and product.
For larger engagements we work with a small network of trusted specialists — designers, regulatory advisors, additional senior engineers — assembled for the specific shape of the work. The aim is the same either way: senior judgment in front of the keyboard, no juniors learning on the client's time.
The studio also runs its own in-house AI infrastructure — on-premise compute capable of handling complex inference workloads — alongside the major cloud providers. This matters for clients whose data sovereignty or regulatory frame rules out sending information to external AI services.
We take on three to four engagements a year. Most run between two and six months. Some become long-running advisory relationships once the initial build has shipped.
How we work
A model that wins on a benchmark and a feature that survives production are two different things. The work earns its keep when it stays running in customers' hands without supervision.
In regulated and high-stakes contexts, the question that matters isn't "how clever can we make it?" — it's "how do we know it's still working tomorrow?" That changes how we design systems from day one.
Postgres before vector DBs. Server-rendered HTML before client frameworks where it fits. Frameworks chosen for the work in front of us, not for the CV. We choose modern only where modernity earns the maintenance cost.
We're not a team to manage. Not a junior to mentor. The work is the work — design, decisions, code, and the documentation that lets it survive after the engagement ends.
Selected work
Medtech · Factory 4.0
Designed and built a production manufacturing system connecting legacy factory machinery to a modern Unified Namespace, with real-time dashboards for the floor. OT/IT integration inside an MDR-compliant medical device environment — where reliability and traceability are the floor, not the feature.
Research · Regulated AI
Research and engineering around how AI inside regulated systems gets verified and re-verified through its lifecycle. Building the scaffolding that lets a medical device manufacturer prove — and keep proving — that the AI inside it is working as documented, even as the model adapts.
Medtech · Compliance
Designed and built a system linking product risk management to training records across a medtech manufacturer — connecting ISO 14971 risk to the ISO 13485 training and competence requirements that follow it. The plumbing a quality team uses to prove, to internal auditors and notified bodies, that the right people have been trained on the right risks for the right products.
Cognitive systems
Designed and built a novel cognitive architecture — putting structured reasoning on top of language models. Research-grade work, taken from a sketch through to a working system.
Enterprise AI
Production integration of Microsoft Copilot with enterprise platforms outside the M365 surface — bringing the assistant into the systems where teams actually work. Hybrid Microsoft SSO, structured data access, and an evaluation harness for where the assistant earns its keep and where humans still close the loop.
Sports
Live club platform — membership, fixtures, communications — built for a working sports club. A reminder that the studio's discipline applies across domains, not just AI.
Work with the studio
Three to four engagements a year, most between two and six months. If the shape of yours fits, you'll hear it in the first call.