Lund is the region’s research centre: the university, ESS, MAX IV, Ideon, and a large number of companies building products on a scientific foundation. That makes AI work different from a conventional industry. The models do not just have to work — the results have to be reproducible, and in many cases the data may not leave the organisation at all.
Cloud Ape AB works with Lund companies from our office in Staffanstorp, twelve minutes away. We translate research and pilots into infrastructure that can actually be operated: GPU capacity, serving layers, data flows, and the traceability that makes an AI system auditable.
What we do for companies in Lund
- GPU infrastructure — bare-metal NVIDIA DGX, GPU scheduling and storage for both training and inference.
- MLOps and AI operations — experiments, model versions and reproducible pipelines instead of notebook files in someone’s home directory.
- LLM serving with vLLM — your own models in production with predictable latency, including fine-tuned variants that cannot be sent to an external API.
- RAG and vector search — Qdrant-backed search over papers, reports, protocols and prior runs, with a citation in every answer.
- Agentic AI — coding and analysis agents that run inside your own environment.
- Kubernetes and platform engineering — a platform a small team can genuinely operate, not a cluster that needs its own department.
From research to production
The most common problem we see at research-heavy companies is not that the model is wrong — it is that the path from “works in a notebook” to “runs in production” was never built.
- Reproducibility. Every run should be recreatable with the same model version, the same data and the same parameters. That matters for internal review, and when a result has to be defended to a customer or an investor.
- Data boundaries. Sensitive datasets stay in your environment. We put inference where the data lives instead of moving the data to the model.
- Evaluation over time. A model that was good in January is not necessarily good in June. We set up continuous evaluations so quality regressions are caught before your users notice.
- Cost per run. GPU hours are expensive. We size for the workload you actually have and make sure idle capacity is not sitting around costing money.
Life science and Medicon Valley
Life science environments demand more traceability than most industries. We build for that:
- Complete logging of prompt, context, model version and output, so an answer can be reconstructed after the fact.
- Access control and isolation between projects, so data from one study cannot leak into another.
- Human in the loop at every step that affects a decision — the AI proposes, a person is accountable.
- Documentation describing the system in plain language: what it does, what data it sees, and how it is monitored.
We do not take on the role of regulatory adviser or validation partner. We make sure the technical side holds up, and that your quality function gets the documentation it needs.
Typical engagements in Lund
- A research company with its own GPU hardware that needs an operating model, scheduling and monitoring instead of manual SSH sessions.
- A life science company that wants to use LLMs internally but cannot send sensitive data to an external provider.
- A startup scaling up from one pilot to a hundred customers, needing infrastructure that holds without the team growing at the same rate.
- A university spin-off with a working prototype that needs help turning it into something that can be operated and sold.
How an engagement runs
- Technical review. We look at the current code, data and infrastructure and assess what can be reused.
- Smallest working production path. One narrow route all the way from data to an answer in production, with metrics in place.
- Handover. Infrastructure as code, documentation and a walkthrough with your team — so you can operate it yourselves.
Frequently asked questions
Are you located in Lund?
Our office is in Staffanstorp, twelve minutes from Lund. We work on site in Lund when needed and remotely for the ongoing work.
Can you help a research company that cannot move its data?
Yes. We set up dedicated GPU capacity and self-hosted inference inside your own environment so sensitive data never leaves your systems. We also build reproducible pipelines so a result can be recreated with exactly the same model and data — even six months later.
We are a small team — does this suit us?
That is often where we add the most value. A small team gets senior expertise part-time instead of a new hire, and a platform designed from the start to be operated by one person.
Do you work with Medicon Valley?
Yes. We deliver to life science companies in the Lund–Malmö region and in greater Copenhagen, focused on traceability, reproducibility and control over where data is stored.
Contact
Cloud Ape AB Skånevägen 26 245 38 Staffanstorp Sweden
Email: info@cloudape.ai Company reg. no.: 559480-8346
Book a call or see all our services. We also work in Staffanstorp and Malmö — and across Skåne.