Summary
Oskar Hallström is an AI research scientist with ~5 years of hands-on experience building and adapting large language and vision models for enterprise deployment, currently advancing reinforcement learning workflows at Adaptive ML. He previously led Training and Inference at LightOn, shipping efficient VLM/LLM adaptations, large-scale pretraining (ModernBERT with 20M+ downloads) and one of the first open RLHF efforts (Alfred-40B) using PPO and custom context-extension techniques. Comfortable across distributed training at supercomputer and multi-node GPU scale, he combines research rigour from Linköping’s Reasoning & Learning Lab with production-focused optimization for customer GPU setups. Multilingual and internationally trained (EPFL exchange, Bocconi and Montréal programs), he pairs deep technical chops in reward modeling, triton/PyTorch kernel work and 3D parallelism with a knack for translating research into enterprise-ready solutions.
5 years of coding experience
2 years of employment as a software developer
Finance Summer Program, Finance Summer Program at Università Bocconi
French Immersion Summer Program, French Immersion Summer Program at Université de Montréal
M Sc in Computer Science / Industrial Engineering and Management - International (French), M Sc in Computer Science / Industrial Engineering and Management - International (French) at Linköping University
Natural Science, Natural Science at The Viktor Rydberg Schools Foundation
Full year exchange at the School of Computer and Communication Sciences Datavetenskap, Full year exchange at the School of Computer and Communication Sciences Datavetenskap at EPFL
Swedish, English, French