Alexandre Carminot

• • Developed novel compositional architecture for de novo phage genome generation achieving 92% re construction accuracy from sparse datasets (850 sequences, 70k vocabulary)

• • Key Innovation: Hierarchical encoding approach that learns functional relationships rather than token level patterns

• • Designed custom Transformer variant with LLM-inspired components for extreme data efficiency and cross-domain transfer

• • Implemented interpretability framework using attention mechanisms for biological validation

Led personalized tutoring sessions in Math and Physics for 10 students, strengthening foundational skills and boosting student confidence in subject mastery.

Led technical workshops on computational methods in astronomy. Organized interdisciplinary talks on AI applications in astrophysics.