Carles Gelada Saez
Carles Gelada Saez
AI Researcher; Co‑founder & CTO at Manifest AI. Former Research Scientist at OpenAI and Brain Resident at Google (research focus: reinforcement learning, representation learning, long-context AI architectures).
Projects
Ideas brought to life. A look at the products, companies, and creative works built from the ground up.
Power Retention (Manifest AI)
A new architecture developed at Manifest AI that replaces attention for long-context modeling, enabling models to handle millions of tokens efficiently (open-source release and accompanying research).
Power Coder (Manifest AI)
Long-context coding autocomplete model released alongside Power Retention (open-source project from Manifest AI).
Brumby-14B-Base (Manifest AI)
Attention-free base model released by Manifest AI demonstrating the capabilities of Power Retention-style architectures.
Vidrial (Manifest AI)
A framework for performant CUDA kernels released by Manifest AI as part of their research engineering toolkit.
DeepMDP: Learning Continuous Latent Space Models for Representation Learning
ICML 2019 paper (co-author) introducing DeepMDP, a method for learning latent-state models useful for planning and representation learning in RL.
Dopamine: A Research Framework for Deep Reinforcement Learning
A flexible research framework for reinforcement learning (co-author on the 2018 preprint/framework).
The Importance of Pessimism in Fixed-Dataset Policy Optimization
ICLR 2021 paper (co-author) on principles for fixed-dataset policy optimization in offline RL.
In the Spotlight
A collection of articles, interviews, and viral content where achievements have been shared.
Recognitions
Formal validation from the top. A showcase of official awards, grants, and government acknowledgements.
O-1A Visa
Aliens of Extraordinary Ability. Reserved for individuals who have risen to the very top of their field.
DeepMDP — ICML 2019 (paper)
2019Highly cited ICML 2019 paper co-authored by Carles Gelada (representation learning / latent dynamics).
Dopamine (research framework) / preprint
2018Research framework and preprint co-authored by Carles Gelada used widely in RL research.
The Importance of Pessimism in Fixed-Dataset Policy Optimization — ICLR 2021
2021ICLR 2021 publication co-authored by Gelada addressing offline RL and pessimism-based techniques.
Google Scholar profile / citation metrics
currentPublic Google Scholar profile listing multiple peer-reviewed publications and citation count (2,200+ citations across works).
Work Experience
Key roles that defined the journey.
Co‑founder & CTO
Technical co‑founder and CTO at Manifest AI; leading research and engineering for long-context AI architectures (Power Retention), open-source model releases (Power Coder, Brumby-14B-Base) and research engineering (Vidrial).
Research Scientist
Research scientist role at OpenAI (research contributions in RL/representation learning; co-author on papers with OpenAI collaborators).
Brain Resident (Google Brain Residency / Research)
Brain Resident / research position at Google Brain (collaborations with Bellemare, Buckman and others; published joint work).
AI Research Intern
AI research internship at Tractable (computer vision / applied ML research).
Membership
Affiliations and memberships in prestigious organizations and communities.
Competitive residency / research program (listed on LinkedIn as 'Brain Resident'); a selective early-career research program at Google Brain.
Education
Academic foundation and learning journey.
No formal education entries are listed on the provided LinkedIn profile and no definitive public education record was found during verification. (If you provide additional information, this section can be expanded.)
Socials
Connect and explore.