Description du poste
*Le descriptif de l’offre ci-dessous est en Anglais*
**Type de contrat :** Stage
**Niveau de diplôme exigé :** Bac + 5 ou équivalent
**Fonction :** Stagiaire de la recherche
**Niveau d'expérience souhaité :** Jeune diplômé
**A propos du centre ou de la direction fonctionnelle**
-------------------------------------------------------
*Created in 2008, the Inria Saclay Center is located at the heart of the Paris-Saclay scientific and technological excellence cluster, which alone accounts for 15% of French research. Serving the development of the Université Paris-Saclay and the Institut Polytechnique de Paris, the Inria Saclay center employs 80 people in research support services and 500 scientists of 54 nationalities.*
*Benefiting from continuous growth, the center now has a total of 42 project-teams and two in the process of being created, including 21 jointly with the Institut Polytechnique de Paris, 16 with the Université Paris-Saclay, as well as 7 Inria EPs, including one in collaboration with Onera and one with the Pôle Universitaire Centre Val de Loire. These research teams are spread over more than ten sites.*
**Contexte et atouts du poste**
-------------------------------
In the context of the ANR FM2AI, and in collaboration with APHP, the aim of this internship is to systematically evaluate LLM-based agents that combine reasoning with specialized medical imaging tools for abdominal CT question answering, focusing on their accuracy, computational efficiency, the contribution of individual tools, and the faithfulness of their reasoning.
*Scientific context:* Foundation models for CT have multiplied: vision‐language encoders (CT‐CLIP [10], Merlin [8], Jolia [15]), generative models that answer questions about a volume (CT‐CHAT [10], MedGemma 1.5 [11]), large pretrained encoders (Pillar‐0 [12], Curia [13], FlexiCT [14]) and segmentation tools (TotalSegmentator [9]). On their own, they still fail on tasks that require localizing, measuring and then reasoning.
Hence the idea, following ReAct [16], of delegating the reasoning to an LLM that calls these models as tools. Born in chest radiography with MedRAX [7], this paradigm now reaches 3D CT: CT‐Agent [2], 3DMedAgent [5], CT‐Flow [6], INFORM‐CT [3].
The most striking result comes from RadAgent [1]: without any training, an agent orchestrating CT‐CHAT, CT‐CLIP, TotalSeg‐mentator and a general‐purpose 2D VLM (Gemma 3 [19]) outperforms CT‐CHAT on report generation. On DeepTumorVQA [4], a fine‐tuned 2D VLM even beats the best 3D model (66.3% vs. 59.8%), but without measurement tools, all models make up quantitative values.
*The problem*: each system comes with its own harness (agent loop, tools, prompts), data and metrics, so it is impossible to tell what actually makes the difference. Abdominal CT, less studied than the chest, is an ideal testbed: DeepTumorVQA pro‐vides questions with verifiable answers on 9,262 public volumes, organized by level (recognition, measurement, reason