CV
Experience
-
2026 - present Lausanne, Switzerland
Graduate Student Researcher
VILAB - EPFL
Continuous Generative Modeling for Unified Any-to-Any Multimodal Systems
- Extending unified any-to-any multimodal architectures via continuous generative processes and flow-based decoders across heterogeneous sensory modalities.
-
2026 - present Lausanne, Switzerland
-
2026 - 2026 Barcelona, Spain
Undergraduate Student Researcher
LAMP - CVC - UAB
Unified Multi-Modal Continual Learning
- Formulated a sequential benchmark to measure cross-modal catastrophic forgetting between visual reasoning and image synthesis in unified multimodal generative models.
- Found that decoupled architectures mitigate bidirectional representation degradation compared with monolithic architectures.
- Awarded the Elena Maseras Research Scholarship 2026 for academic excellence (13% acceptance rate).
-
2026 - 2026 Barcelona, Spain
Undergraduate Student Researcher
IRI - UPC
Future Action Prediction in Video Models
- Developed WarmFlow, an encoder-initialized flow-matching framework that replaces pure-noise diffusion with dynamic structural priors for action segmentation.
- Achieved competitive generative coherence with one function evaluation (NFE=1), reducing computation by 10x-25x for latency-critical applications.
-
2025 - 2026 Lausanne, Switzerland
Visiting Student Researcher
BioRob - EPFL
Context-Aware Spatial Action Recognition
- Extended the SkateFormer architecture for multimodal spatial action recognition, using geometric priors to generalize to occluded layouts.
- Curated a 10k-sample dataset and deployed the framework on a ROS2 mobile swarm, achieving a 20% accuracy gain over baselines and 25ms latency for closed-loop control.
-
2025 - 2025 Barcelona, Spain
AI Engineer Intern
PayRetailers
Multi-Agent Systems and LLM Analysis
- Evaluated prompting strategies, including chain-of-thought and few-shot prompting, and connected LLMs to external tools through multi-agent workflows.
- Built internal B2B AI solutions, including retrieval-augmented generation systems and automated analytical pipelines.
Education
-
2026 - present Lausanne, Switzerland
-
2022 - 2026 Barcelona, Spain
-
2025 - 2026 Lausanne, Switzerland
Publications
-
Continual Adaptation Across Modalities: Learning and Forgetting in Unified Multimodal Models
NeurIPS 2026 Workshop on CL4FMAgents
Xinrui Wang , Zhihao Chen , Jia Tang , Alexandra Gomez-Villa , Bartłomiej Twardowski , Joost van de Weijer , Songcan Chen
Projects
-
B2Travel - HackUPC 2026
Second Prize, Skyscanner Challenge; Best Use Award of ElevenLabs.
Built an LLM-grounded multimodal recommendation system that maps CLIP visual-semantic embeddings into a 3D topological space for spatial retrieval.
-
MXEgg - LauzHack 2025
Overall finalist, LauzHack 2025.
Built an LLM-driven multi-agent simulation with a custom physics engine and localized vision to study emergent cooperative behavior.
-
AIvidence - MERIThon 2025 AI Challenge
First Prize, MERIThon 2025 AI Challenge.
Built a retrieval-grounded multi-agent framework that distributes chain-of-thought reasoning across specialized LLMs for deepfake detection.
Awards
-
Elena Maseras Research Scholarship 2026
LAMP - CVC - UAB
Awarded by LAMP, CVC-UAB for academic excellence (13% acceptance rate).
-
Santander Excellence Scholarship
Santander 360º
An award recognizing top academic performance as well as participation in cultural, social, and professional development activities (only 77 recipients selected across all Spanish universities and academic fields).
-
Overall Finalist, LauzHack 2025
EPFL, Switzerland
-
First Prize, MERIThon 2025 AI Challenge
UPC, Spain
-
Academic Honors
UPC, Spain
Four courses with distinction.