I take AI from research prototype to production: Computer Vision, medical AI, and LLM systems for teams whose problems don't fit standard tools — delivered as tested, maintainable software.
Intro call ≈ 30 minutes · reply within 24–48 hours · no commitment
Expertise
Grouped by problem, not by tool. Behind every area are systems I have designed, implemented, and evaluated — in research and in engineering projects.
Built hyperspectral imaging workflows covering calibration, spatial alignment, spectral preprocessing, dimensionality reduction, and AI-assisted analysis. Developed deep-learning pipelines for large medical image datasets, including whole-slide processing, patch extraction, and tumor analysis.
Designed and evaluated methods that use LLMs to transform categorical data into meaningful numerical representations. Built repeated evaluation pipelines, embeddings, semantic similarity analysis, and RAG-based workflows.
Research in visual analytics, similarity visualization, and AI-assisted visualization, published in leading venues. I make high-dimensional and spatio-temporal data interpretable for domain experts.
Built reproducible pipelines for large datasets, SQL-backed systems, batch processing, and GPU and HPC workflows, including optimization work on roughly 30 GB of MS SQL data.
Robust Python applications with testing, containers, CI/CD, and REST APIs. I turn research prototypes into maintainable, tested systems.
Engagement Models
I take on a small number of projects with research-driven teams — biotech and medical-device companies, scientific R&D labs, and engineering teams working with high-dimensional data. Each engagement is scoped around the problem to solve, not around a fixed toolkit.
A focused scoping engagement: architecture review, model feasibility, GPU and data-pipeline bottleneck analysis, and data strategy. You get an architecture recommendation with an honest go/no-go and the reasoning behind it — before you commit a full development budget.
Embedded delivery of the system itself — Computer Vision pipelines, LLM and RAG applications, scientific software. From first prototype through evaluation to a tested, documented handover.
Fractional technical-lead support for internal R&D teams: refining algorithms, establishing evaluation pipelines, and hardening research code for production.
Selected Work
These case studies show how I approach AI problems that lack an obvious solution — and what remains at the end: working, maintainable software. Where measurable results are not yet published, I describe the state precisely instead of inventing numbers.
H&E · pituitary whole-slide image
02
Research
Research is not a side note here — it is where my engineering standards come from. Peer review forces precise problem statements, honest evaluation against baselines, and reproducible results. I hold client projects to the same bar.
Selected publications
Technologies
The problem comes first; these are the tools I reach for to solve it. For technical readers who want the details.
How I work
Structured, transparent, and hypothesis-driven. You always know where a project stands and why.
Clarify the real goal, constraints, and data before proposing any method.
Test the hardest assumption early and state honestly what is and is not achievable.
Develop the model or system hypothesis-driven, with evaluation built in from the start.
Turn the validated approach into reliable, maintainable software that fits your stack.
Explain methods, limitations, and results clearly to technical and non-technical stakeholders.
Why work with me
I take ownership from problem definition and architecture through implementation, evaluation, and handover. My background in research, teaching, and interdisciplinary development helps me communicate complex technical decisions clearly.
Research-level depth: when no proven method exists, I can design and validate one
Evidence-driven: hypotheses, baselines, and evaluation pipelines instead of demos that work only once
Honest feasibility assessments — including “this will not work, and here is why”
Ambiguity is my working terrain: vague research questions become scoped, testable engineering plans
Interdisciplinary communication: physicians, researchers, engineers, and decision-makers get the same clear story
Maintainable by default: tested, documented, containerized software your team can extend without me
Key information for recruiters
A compact overview for recruiters, agencies, and technical decision-makers.
Need my CV? I'd be happy to share my latest CV together with additional background information.
Request my CVAbout
Karim Huesmann · Münster, Germany
I work at the intersection of research and practical software engineering. My background spans physics, economics, computer science, data analysis, and scientific visualization, an interdisciplinary path that shapes how I approach problems.
I structure complex data problems, develop suitable AI methods, and prepare results so they are reliable and understandable. I move quickly into new domains and collaborate closely with researchers, engineers, medical experts, and decision-makers.
The problems I enjoy most are the ones without an obvious solution — where the method has to be designed, not just applied. Alongside client work I am completing a doctorate in visualization at the University of Münster, where I have taught and supervised students. That work keeps me close to how ideas become reliable systems.
Path
Trust & links
FAQ
Computer Vision and imaging systems, including hyperspectral and medical whole-slide pipelines, LLM and RAG applications with evaluation pipelines, and reproducible scientific software for data-intensive problems. Engagements range from prototypes to productization and technical consulting.
Yes. I work remote-first from Münster, Germany, with occasional on-site workshops or kick-offs across Europe.
Yes. I have built image-analysis pipelines covering calibration, preprocessing, segmentation, representation learning, and deep-learning classification for scientific and medical data, including hyperspectral imaging and whole-slide image processing.
Yes. I design grounded LLM workflows with embeddings, semantic similarity, prompt engineering, and RAG, and I build repeated evaluation pipelines that measure them against established baselines.
Yes. I am an active researcher with peer-reviewed publications in IEEE TVCG, Computer Graphics Forum, and IEEE CG&A, and I have worked in interdisciplinary teams across computer science and medicine.
I start with the problem and the data, test the hardest assumption early, and give an honest feasibility assessment. Development is hypothesis-driven with evaluation built in, and results are handed over as tested, maintainable software.
Yes, for selected projects from Q4 2026. For larger engagements it helps to reach out early.
Contact
Describe it in a few sentences — no pitch deck needed. You will get a personal, honest assessment of whether and how I can help. And if I am not the right person for it, I will tell you that too. You talk directly to me — there is no team and no sales layer behind this form.