About this role
Fixed term contract | Belval | 24 Months Are you passionate about research? So are we! Come and join us The Luxembourg Institute of Science and Technology (LIST) is a leading Research and Technology Organisation (RTO) that drives innovation for the economy and society in Luxembourg and beyond. With cutting-edge expertise in Natural, Built, Industrial environments, Space, AI, Security and defence technologies. LIST bridges scientific excellence and applied research to design solutions that address real-world challenges and create positive impact. Do you want to know more about LIST? Check our website: https://www.list.lu / How will you contribute? You'll be working on a production-grade Retrieval-Augmented Generation (RAG) system purpose-built for second-level controls on NAV computation in investment funds. The system must retrieve relevant regulatory methodology, replicate calculation steps, detect discrepancies against fund accounting outputs, and produce control reports that are traceable, auditable, and defensible under regulatory scrutiny. You will contribute to a system where precision, explainability, and reliability are non-negotiable. Code Development (60%): · Build retrieval pipelines optimised for structured regulatory methodology documents (CSSF circulars, ESMA guidelines, fund prospectuses), with adaptive weighting to surface the specific calculation rules relevant to each control query · Implement numerical validation pipelines that compare LLM-derived NAV calculation outputs against fund administrator figures, applying configurable tolerance bands and discrepancy flagging logic · Implement chunking strategies optimised for fund accounting reports and regulatory circulars, preserving the structure of calculation tables, formula references, and numbered regulatory provisions. · Build evaluation frameworks measuring control accuracy (error detection rate, false positive/negative rates on flagged discrepancies) and the quality of generated audit reports · Optimise vector database indexing for regulatory document corpora with high update frequency, ensuring fast retrieval as new CSSF/ESMA guidance is published Research & experimentation (25%): · Benchmark embedding models against the current Nomic Embed Text baseline, with a focus on sensitivity to numerical and regulatory language in fund accounting contexts · Explore cross-lingual retrieval for EU regulatory documents published across multiple languages, with emphasis on French and German variants of CSSF and ESMA guidance · Research agent-based simulation of NAV calculation steps, investigating how LLMs can reason reliably over arithmetic-sensitive regulatory workflows and signal confidence levels · Experiment with multi-agent orchestration for second-level control workflows, where agents decompose a NAV control task into parallel verification sub-tasks across pricing, accruals, and corporate actions. System integration (15%): · Integrate LLM backends optimised for structured reasoning and develop prompt engineering patterns that enforce step-by-step calculation verification and explainable output · Implement comprehensive logging, structured audit trail generation, and monitoring dashboards that allow compliance officers to review and challenge control decisions Is Your profile described below? Are you our future colleague? Apply now! Education · A PhD in Computer Science, Artificial Intelligence, or a closely related field is required. · A background in Financial services domain is highly desirable. Experience and skills · Preferably at least 2 years of research or industry experience in relevant areas. Must Have: · Strong Python programming skills · Hands on in LLMs RAG implementation . · Experience with at least one ML framework (scikit-learn, transformers, etc.) · Familiarity with data structures and algorithms · Git proficiency for collaborative development Highly valued: · Previous work with LangChain, LlamaIndex, or similar retrieval and agent orchestration frameworks, ideally in a compliance or audit context · Knowledge of multi agent workflow orchestration frameworks. · Experience with NLP libraries (spaCy, NLTK, transformers) applied to structured extraction from regulatory or financial reporting documents · Understanding of information retrieval concepts (BM25, TF-IDF, relevance scoring) · Experience with React, Node.js or similar web frameworks · Knowledge of evaluation methodologies for ML systems · Track record of conducting application-driven AI research. · Ability to work independently as well as collaboratively in interdisciplinary and multi-partner research environments. · Understanding of NAV calculation methodology, fund accounting principles, or second-level control frameworks in asset management · Familiarity with CSSF or ESMA regulatory frameworks governing fund operations, pricing, and oversight obligations Language skills • Fluency in English (and French), both oral and written. • Other relevant languages are an asset. Your LIST benefits · An organization with a passion for impact and strong RDI partnerships in Luxembourg and Europe that works on responsible and independent research projects · Sustainable by design, empowering our belief that we play an essential role in paving the way to a green society · Innovative infrastructures and exceptional labs occupying more than 5,000 square metres, including innovations in all that we do · An environment encouraging curiosity, innovation and entrepreneurship in all areas · Personalized learning programme to foster our staff's soft and technical skills · Multicultural and international work environment with more th