AI systems, engineered
for real use.

AI & software engineer·researcher·Essen, Germany

AI and software engineer, writing software professionally since 2015. Builds language-model systems that run outside the lab, mostly in medicine, sometimes in games. NLP team lead at the Institute for AI in Medicine (IKIM) in Essen; works with other teams through Aitecture UG.

Get in touchResearch
Osman Alperen Çinar-Koraş
osman alperen çinar-koraş
ai engineer · essen, de
available for consulting
00
[01]

What I work on

AI in healthcare

Language-model systems for a university hospital, built on live patient data together with clinicians: registry extraction from patient records and a writing assistant for discharge letters. All on-premise. The extraction system is in use at University Medicine Essen (EMNLP 2026 Industry Track).
clinical-nlpfhirinformation-extractionon-premise

Agents and LLM systems

Agent harnesses for concrete jobs: collaborative writing, information extraction, document generation. Tools, retrieval and evaluation, with a human in the loop.
agentsraggroundingtool-usemcpevaluation

NLP and model training

NLP with a focus on language models: fine-tuning open models, synthetic training data, text generation, summarisation and evaluation. A new state of the art on the BioNLP Discharge Me! task (ACL 2025 Findings). → Research
fine-tuninglorasynthetic-datasummarisationpytorch

Running models in production

Serving the hospital's in-house models on Kubernetes and its own GPUs: language models, OCR, layout detection and streaming speech recognition.
vllmkubernetesgpu-servingocrspeech-recognition

Simulations, games and digital humans

Unreal Engine 5:
  • environments for intelligent agents, such as simulations of people;
  • semi-procedural generation, with agents working on mathematical models;
  • conversational digital humans with long-term memory;
  • two engine plugins on Fab, for language models and structured output.
unreal-engine-5c++metahumanprocedural-generationagent-simulation

Software engineering

Since 2015: university web and desktop applications, Office extensions for financial reporting, and today the backends, APIs and collaborative editors around AI systems.
pythontypescriptc#c++rustreactvuepostgresdocker
[02]

Research

Publications on clinical text generation, information extraction and evaluation, at EMNLP, ACL and ECIR.

Full research path →
[03]

Path

From mathematics and software engineering to AI research and an NLP team at a university hospital.

  1. NLP Team Lead · IKIM, Essen2025 – today
  2. AI researcher (PhD) · IKIM, Essen2022 – today
  3. Software engineer · firesys, Frankfurt2020 – 2022
  4. Software developer (part-time) · JGU Mainz2015 – 2018
  5. Mathematics, B.Sc. & M.Sc. · JGU Mainz2013 – 2021
[04] working together · aitecture ug

Working together

Through Aitecture UG, I work with startups and established companies that build on language models. My advice comes from building and shipping these systems myself.

specialty · healthcare

Clinical AI is where mistakes cost the most. Since 2022, I build AI systems on clinical data at University Hospital Essen: real patient records, strict privacy requirements, and clinicians who have to be able to trust every output. One of them is in clinical use today.

Usually starts with a short email:[email protected] →
  1. [a]

    Technical Assessment

    An independent second opinionof an AI feature, a vendor or an architecture: what holds up, what doesn't, and what to do about it.

  2. [b]

    Feasibility & Architecture

    Whether a use case works on your data, and how it should be built. With a small prototype where that answers the question faster.

  3. [c]

    Build & Deploy

    Agents, retrieval pipelines, information extraction and fine-tunes, built with your team and taken into real use.

  4. [d]

    Evaluation

    Test sets, expert feedback loops and error analysis, so you can measure an LLM system and know what to fix next.