Timothée Coste

Collaborated in an international team [working language English] of data scientists to build and deploy a distributed data processing pipeline: . Using proprietary and open source AI models to extract metadata from heterogeneous files and maintain a data catalog . Modeling data scheme for efficient querying . Resulting in indexing 100k+ semantically rich documents, accelerating data exploration and retrieval and unveiling data labelling issues

Contributed to HybridAGI, an LLM-based neuro-symbolic AI system: benchmarked five state-of-the-art LLMs and developed prompts for synthetic data generation with a self-refinement loop.

xAI: Studied with co-intern counterfactual reasoning properties in constrained binary classifiers and contributed to the implementation of two case studies.