Open to new roles from January 2027

AI Researcher & Project Lead

Kristian Noullet M.Sc.

Tech enthusiast. AI realist.

NLP · Large Language Models · Semantic Web · AI Project Leadership

I build AI systems that turn unstructured text into connected knowledge, and I lead the projects that bring them into real organisations. Doctorate at KIT (defense on 10 December 2026), seven years of applied AI projects with partners such as BMW, DATEV, Holtzbrinck and Nokia.

Get in touch See my projects
  • 5AI projects with industry partners
  • 3of them as project lead
  • €215k+project budget managed
  • up to 25student workers per semester
  • 7publications, incl. IJCNLP-AACL & K-CAP
  • 5languages at C2 or mother-tongue level

About me

From research to practice

I'm a computer scientist specialising in natural language processing, large language models and the Semantic Web. During my doctorate at the Karlsruhe Institute of Technology (KIT), I developed modular, knowledge-graph-agnostic methods for entity linking (recognising which real-world thing a word in a text refers to) and released them as open-source frameworks.

Alongside research, I spent about seven years leading and delivering applied AI projects with industry partners: from LLM-based policy compliance checking at DATEV to a network-planning tool in use at atesio. Before that, I worked as a full-stack developer at ING Luxembourg, where my tools replaced commercial software costing EUR 150k per year in licences.

I enjoy bringing technology, business units and management together. For me, a project is only finished when its results are actually used. I'm now looking for a role where I can lead AI projects like that.

Career

Professional experience

  1. 09/2019 - 08/2026Karlsruhe

    Research Associate (Doctoral Researcher)

    Karlsruhe Institute of Technology (KIT), Institute AIFB, Web Science group

    • 5 AI projects, 3 as project lead: delivered with industry and consortium partners (incl. BMW, DATEV, Holtzbrinck, Nokia), with team and budget responsibility totalling over EUR 215k.
    • Entity linking research: developed methods that use transformers and LLMs to automatically link terms in text to the matching entries in knowledge bases. They work with any data source, from corporate wikis to Wikidata, and were released as the open-source frameworks CoLiT and Agnus LLM and published at international conferences such as IJCNLP-AACL.
    • Agnus LLM: 3.7 % better than the state of the art on average across 5 benchmarks, using language models 100 times smaller than the weaker competing LLMs. Prevents hallucinations, eliminates 15.2 % variability in results (F1) and uses synthetic test data to measure whether models have memorised benchmarks.
    • Teaching leadership: up to 900 students per semester across my courses, incl. up to 800 in the tutorial programme for Foundations of Computer Science I/II; led up to 25 student workers (tutors and project assistants); designed and taught my own Master's course Knowledge Discovery, Graph Neural Networks, and Language Models; supervised 7 final theses.
    • Funding: (co-)authored grant proposals, e.g. the renewal proposal for the DFG research unit FOR 5339 (multi-year funding for 2-3 doctoral researchers).
    • Institute web portal: principal developer and administrator of the AIFB portal (approx. 30,000 pages; Semantic MediaWiki, Docker, Squid); ran, maintained and migrated it to new systems.
    • Python
    • PyTorch
    • Hugging Face
    • LLMs
    • Ollama
    • Docker
    • Semantic MediaWiki
  2. 02/2018 - 08/2019Freiburg

    Research Assistant (part-time)

    University of Freiburg, Chair of Databases and Information Systems (Prof. Dr. Georg Lausen)

    • Research on NLP, relation extraction, the Semantic Web and machine learning; built proofs of concept.
  3. 02/2018 - 08/2019Freiburg

    Systems Administrator (part-time)

    University College Freiburg, University of Freiburg

    • Ran IT operations for more than 50 employees in a team of two: administered a heterogeneous Windows, Linux and macOS landscape.
    • Managed users and access rights via Active Directory and Group Policies; set up accounts and workstations for new employees.
    • Operated network infrastructure, DHCP and firewalls; automated recurring administration tasks with PowerShell scripts.
    • Windows
    • Linux
    • macOS
    • Active Directory
    • PowerShell
  4. 02/2014 - 08/2015Luxembourg

    Full-Stack Software Developer

    ING Luxembourg S.A. (intern until 08/2014)

    • Built a software toolkit for the IT department (approx. 120 staff) with Java, Spring, SQL and JavaScript, incl. a request management system and a development environment for COBOL: replaced commercial software with licence costs of EUR 150k per year, with more functionality.
    • Fully automated mainframe workflows (TN3270 connectivity to IBM mainframes, reverse engineering of undocumented mainframe behaviour): screen sequences that took up to an hour of manual clicking now run without user input, saving several hundred developer hours per month.
    • Java
    • Spring
    • SQL
    • JavaScript
    • IBM Mainframe

Selected work

Projects

Applied AI projects with industry and consortium partners: goal, approach and result.

01/2025 - 08/2026

LLM-GALP: LLMs for policy management and AI governance

Software Campus project with DATEV eG

  • Project lead
  • Budget EUR 115k
  • EU AI Act
Goal
Use large language models to analyse and manage internal policies and AI governance requirements (EU AI Act) automatically for business decisions.
Approach
Took over and completed the project; built LLM-based policy compliance checking with locally hosted open-weight LLMs (Python, Docker, Ollama).
Result
Prototype deployed internally; automates large-scale policy analysis, saves users time and supports compliance checks and EU AI Act compliance decisions.

06/2022 - 06/2024

FWM (Find What Matters): detecting key content

Software Campus project with Holtzbrinck Publishing Group

  • Project lead
  • Budget EUR 100k
  • Team of 4 assistants
Goal
Make scientific publications faster to grasp by detecting their key content automatically.
Approach
Transformer-based methods (encoder and decoder models) combining several sources of information (summary, keywords, context, method and results along the IMRaD structure); technical lead of 4 student research assistants.
Result
Piloted with Holtzbrinck on scientific publications. Research results: the open-source frameworks CoLiT and Agnus LLM, plus the basis for "Linking with Bias" (SEMANTiCS 2025) and the follow-up ReCoLTe (Recommending Combinations of Linking Techniques, in preparation).

2020 - 2024

KIGLIS: AI for optimising fibre-optic networks in smart cities

Joint project with Nokia, VPI Photonics, Telemaxx, KIT-IPQ and atesio

  • Work package lead
  • Consortium of 15 people
  • Multi-million budget
Goal
Use AI to expand fibre-optic networks in smart cities more intelligently and to steer traffic better.
Approach
Developed ML methods on infrastructure and traffic data; passed on AI know-how to the industry partners in regular workshops.
Result
Tool built for the partner atesio and in use there: improved the location accuracy of infrastructure maps from hundreds of metres to a few metres or even centimetres, which substantially cuts excavation costs for repairs and network expansion.

2019 - 2022

HUBRAUM: making internal corporate knowledge usable with NLP and ML

Research project with the BMW Group

  • Team lead (3 people)
Goal
Make internal corporate knowledge easier to find and use.
Approach
Developed methods that automatically interlink internal knowledge sources, independent of the knowledge base in use.
Result
Prototype for an innovation unit of the BMW Group that semantically links documents and projects into a connected overview of ongoing activities for researchers and management.

2020 - 2022

IIDI: intelligent information services for domain-specific web portals

Joint project with TecIntelli

Developed ML methods for an intelligent, domain-adaptive and continuously improving system that provides contextual information on specialised topics.

2019 - 2026

Open-source frameworks CoLiT & Agnus LLM

GitHub: CoLiT · Agnus LLM

CoLiT combines entity linking systems and components into custom pipelines; Agnus LLM performs entity disambiguation with interchangeable open-weight LLMs.

Research & teaching

Research

Doctoral thesis

Modular and Adaptive Named Entity Linking

Karlsruhe Institute of Technology · submitted 07/2026 · defense on 10 Dec 2026 · supervisor: Prof. Dr. York Sure-Vetter · second reviewer: Dr.-Ing. Tobias Käfer

Entity linking connects mentions in text (such as "Paris", "Jaguar" or "Mercury") to the right entry in a knowledge base. My thesis makes this process modular (systems and components can be combined and swapped), knowledge-graph-agnostic (it works with Wikidata, DBpedia or a company's own knowledge base) and adaptive (robust LLM-based disambiguation that avoids hallucinations).

Publications

  1. 2025

    Agnus LLM: Robust and Flexible Entity Disambiguation with Decoder-only Language Models

    K. Noullet, A. Ourgani, N. T. Lakner, L. Kinder, T. Käfer

    Proceedings of IJCNLP-AACL 2025, pp. 3266-3284, ACL

  2. 2025

    Linking with Bias: Domain-Specific Behaviour in Entity Linking Systems

    K. Noullet, A. Ourgani, N. Lakner, T. Käfer

    Proceedings of SEMANTiCS 2025, Studies on the Semantic Web, IOS Press

  3. 2023

    A Full-Fledged Framework for Combining Entity Linking Systems and Components

    K. Noullet, A. Ourgani, M. Färber

    Proceedings of the 12th Knowledge Capture Conference (K-CAP), pp. 148-156, ACM

  4. 2022

    Which Publications' Metadata Are in Which Bibliographic Databases? A System for Exploration

    M. Färber, C. Braun, N. Popovic, T. Saier, K. Noullet

    BIR Workshop at ECIR 2022, pp. 39-44, CEUR-WS

  5. 2021

    Combining Linking Techniques for Everyone

    K. Noullet, S. Printz, M. Färber

    The Semantic Web: ESWC 2021 Satellite Events, LNCS, pp. 88-92, Springer

  6. 2020

    KG-agnostic Entity Linking Orchestration

    K. Noullet

    Doctoral Consortium at ISWC 2020, pp. 41-48, CEUR-WS

  7. 2020

    KORE 50DYWC: An Evaluation Data Set for Entity Linking Based on DBpedia, YAGO, Wikidata, and Crunchbase

    K. Noullet, R. Mix, M. Färber

    Proceedings of LREC 2020, pp. 2389-2395, ELRA

Full record on ORCID →

Academic service

Co-organiser of SEMANTiCS 2019 (Karlsruhe); reviewer for ISWC, ESWC, WWW, IJCNLP-AACL and others.

Teaching & supervision

  • Summer 2020 - Summer 2026

    Tutorial lead, "Foundations of Computer Science I"

    KIT · Bachelor's programmes in Information Systems and Industrial Engineering

    600-800 students per semester; coordinated 20-23 tutors; created slides and exams; gave lectures.

  • Winter 2019/20 - Winter 2024/25

    Tutorial lead, "Foundations of Computer Science II"

    KIT · Bachelor's programmes in Information Systems and Industrial Engineering

    600-800 students per semester; coordinated 20-23 tutors; redesigned all slides from scratch; created exams; gave lectures.

  • Winter 2022/23 - Winter 2025/26

    Master's lecture "Knowledge Discovery, Graph Neural Networks, and Language Models"

    KIT · Master's level

    Sole instructor for approx. 30 students: LLMs, Semantic Web, machine learning; created all slides and designed the exam.

  • 2022 - 2026

    Seminar "Linked Data and the Semantic Web"

    KIT

  • 2019 - 2026

    Thesis supervision

    7 Bachelor's and Master's theses in NLP, entity linking, LLMs and the Semantic Web.

Education

Education & training

  1. 09/2019 - 12/2026Karlsruhe

    Doctorate (Dr.-Ing., equivalent to Ph.D.)

    Karlsruhe Institute of Technology (KIT), Institute AIFB

    Thesis: "Modular and Adaptive Named Entity Linking", submitted 07/2026, defense on 10 Dec 2026.

  2. 2015 - 2019Freiburg

    Master of Science in Computer Science

    University of Freiburg

    • Final grade: 1.5 (German scale, 1.0 = best)
    • Focus: machine learning, Semantic Web, entity linking
    • Master's thesis: "Agnos - Knowledge Graph-Agnostic Entity Linking"
  3. 2011 - 2014Luxembourg

    Professional Bachelor in Computer Science

    University of Luxembourg

    • Bachelor's thesis: "Paprika - Smart Request Management & Development" (16.5/20)
    • Erasmus semester at Linköping University (Sweden)
    • Scholarship for the Asian Studies summer school at Sophia University, Tokyo (2013)
  4. 2022 - 2026Professional development

    Software Campus: leadership programme for future IT executives

    Funded by the German Federal Ministry of Research, with industry partners such as DATEV and Holtzbrinck

    Trainings: new as a leader; persuading under resistance; recognising and using communication potential; design thinking; change management; diversity & inclusion (leading diverse teams successfully).

Toolbox

Skills

AI, ML & NLP

  • Large Language Models
  • Ollama
  • LangChain
  • LlamaIndex
  • PyTorch
  • Hugging Face Transformers
  • spaCy
  • Named entity recognition
  • Entity linking

Semantic Web

  • Knowledge graphs
  • RDF
  • OWL
  • SPARQL
  • Wikidata
  • DBpedia
  • Semantic MediaWiki

Programming

  • Python
  • Java
  • SQL
  • JavaScript
  • C/C++

advanced

Software & DevOps

  • Spring
  • React
  • Angular
  • REST APIs
  • Docker
  • Git
  • Linux
  • System administration (Windows, Active Directory, networking)

Management

  • Project management
  • Grant proposals
  • Partner management
  • Team leadership
  • Workshops
  • Design Thinking
  • Change Management
  • AI governance (EU AI Act)

Languages

  • EnglishC2
  • GermanC2
  • Frenchmother tongue
  • Luxembourgishmother tongue
  • Finnishmother tongue

Beyond work

Outside the office

Where I come from, where I volunteer and what I do when I'm not training models or teaching.

Roots & languages

I'm a Luxembourgish and Finnish citizen and at home in five languages. I studied in Luxembourg, Sweden and Germany and spent a scholarship summer at Sophia University in Tokyo. Switching between cultures and languages comes naturally to me, which helps when bringing very different teams together.

Community & volunteering

  • Erasmus Student Network, Luxembourg (2013-2019) and Karlsruhe (2019-2024): organising events; National Representative for Luxembourg at EU level, incl. in Brussels.
  • Doctoral Researchers' Council of the KIT Department of Economics and Management (2020-2026), board member 2023-2025.
  • QueerKAstle: supporting events (2024-2026).
  • FabLab Karlsruhe makerspace: volunteer since 2025.

Maker at heart

I enjoy working with my hands: sewing, electronics, woodworking, and cooking and baking traditional food from different countries, from Finnish Janssonin kiusaus, pulla, korvapuusti and karjalanpiirakka to French gratins. To balance out screen time, I also play tennis and go to the gym.

Alongside my personal projects, I'm building a proof of concept for a startup idea focused on trustworthy AI.

  • Home automation
  • Making & crafting
  • Tennis
  • Gym
  • Optimising everyday processes

Contact

Let's talk

I'm open to roles in AI and NLP engineering, applied research and AI project leadership. The best way to reach me is a message on LinkedIn. I'm happy to share my full CV on request.

Message me on LinkedIn GitHub ORCID