Internship In ai & Llm Adaptation Of What-If Simulations For Business Processes H/F - 38

Le CEA est un acteur majeur de la recherche, au service des citoyens, de l'économie et de l'Etat.

Il apporte des solutions concrètes à leurs besoins dans quatre domaines principaux : transition énergétique, transition numérique, technologies pour la médecine du futur, défense et sécurité sur un socle de recherche fondamentale. Le CEA s'engage depuis plus de 75 ans au service de la souveraineté scientifique, technologique et industrielle de la France et de l'Europe pour un présent et un avenir mieux maîtrisés et plus sûrs.

Implanté au coeur des territoires équipés de très grandes infrastructures de recherche, le CEA dispose d'un large éventail de partenaires académiques et industriels en France, en Europe et à l'international.

Les 20 000 collaboratrices et collaborateurs du CEA partagent trois valeurs fondamentales :

- La conscience des responsabilités
- La coopération
- La curiositéBusiness Process Simulations (BPS) are essential for analyzing, optimizing, and forecasting organizational workflows. What-if simulations enhance BPS by enabling analysts to test hypothetical changes-such as new tasks, altered decision points, or alternative flows-before implementation. A key challenge is to set correct simulation parameters for these changes, including task durations, distributions over decisions, and who would execute the involved tasks and when.
For example, a bank may introduce overnight delivery for newly issued credit cards, alongside the existing standard mail and express courier options. To simulate this scenario accurately, we must estimate how often customers will choose the new option and the duration of its associated tasks, like expedited printing, packaging, and courier handoff.
This project aims to develop adaptive methods that realistically model human behavior and decision-making for such what-if scenarios. For this purpose, we investigate the following approaches:
AI techniques similar to Retrieval-Augmented Generation (RAG) using low-dimensional text embeddings and clustering to identify historical process flows similar to new what-if scenarios. The scenario is encoded into a semantic vector, which is then compared to past cases using similarity metrics like cosine distance. Once similar instances are retrieved, their task durations and decision probabilities are extracted and used to estimate the unknown parameters in the new scenario
Large Language Models (LLMs) are able to predict human decisions and responses. Our goal is to use foundation models such as Mistral to directly predict task durations or decision distributions for what-if scenarios. The approach involves optimizing prompt strategies and fine-tuning the models to ensure context-aware, plausible outputs that are aligned with historical and expert knowledge.
Your task:
Perform a literature review about the current state-of-the-art in what-if simulations for BPS and their adaptation
Develop together with your supervisor ideas for new adaptation mechanisms that are based on historical data and/or LLM suggestions on how humans might behave
Evaluate your methods by comparing them to baselines and performing ablation studies
Write a scientific report about your findings, preferably as a scientific workshop or conference contribution
The internship will be carried out at CEA Grenoble.