Federated Learning
Sovereign training across distributed nodes without centralising sensitive data. We are advancing the theory and practice of federated optimisation at the scale required for frontier models.
AICONSORTIUM · Research Division
At the frontier of sovereign AI research
We do not follow the frontier. We help define it. Our research teams work on the hardest open problems in AI, from federated architectures to multilingual language understanding to alignment for diverse societies.
Open Research
6 Active Areas
4 Publications
Open research
We publish what we learn
From federated mixture-of-experts to multilingual language coverage, our work is shared openly. Sovereign capability and open science are not in tension.
Collaborate with us
Active Research
Six frontiers. One team.
Mixture of Experts
Sparse routing at trillion-parameter scale. Our Hierarchically Detached MoE architecture allows expert clusters to operate semi-autonomously, dramatically reducing cross-cluster communication overhead.
Multilingual NLP
Deep language understanding across 100+ world languages, with special depth in underserved and low-resource languages. Morphological complexity, code-switching, script diversity, and oral language patterns: we study them all at research depth.
Legal AI
Jurisdiction-aware reasoning systems built for the complexity of real-world jurisprudence across jurisdictions. Constitutional law, civil and criminal codes, sub-national variation, and multilingual judicial records, all within a single reasoning framework.
Multimodal Reasoning
Cross-modal understanding that goes beyond caption matching. We study how models integrate visual, auditory, and textual signals into unified representations capable of genuine cross-modal inference.
AI Safety
Alignment and constitutional AI adapted for diverse global contexts. We study how value alignment works across societies this diverse, and how to build models that are genuinely safe at the cultural level, not just the technical one.
Publications
What we have published
Our research is open. We publish because the problems we are working on are important to everyone, not just to AICONSORTIUM.
Architecture · 2025
Hierarchically Detached Federated MoE: Architecture and Training
AICONSORTIUM Research Team
We introduce Hierarchically Detached Federated MoE, a sparse mixture-of-experts architecture designed for sovereign deployment. Expert clusters operate semi-autonomously across federated nodes, reducing cross-cluster communication overhead by 94% while maintaining convergence guarantees.
Legal AI · 2025
Sattam: Jurisdiction-Aware Legal Reasoning Across Jurisdictions
AICONSORTIUM Research Team
We describe the design and evaluation of Sattam.ai, a legal reasoning model trained on statutory, constitutional, and case-law corpora across multiple jurisdictions. We introduce a novel benchmark for jurisdiction-aware legal inference spanning dozens of sub-national jurisdictions.
Distributed Inference · 2026
Federated Mixture-of-Experts: Sovereign Distributed Inference Across Compute Boundaries
AICONSORTIUM Research Team
We present the Federated Mesh, a distributed inference architecture that lets a single Mixture-of-Experts model reason across geographically sovereign compute nodes without any node seeing the full context. We demonstrate coherent cross-node synthesis at scale while preserving strict data-jurisdiction boundaries.
Multilingual NLP · 2026
Multilingual Coverage in Large-Scale Multimodal Models
AICONSORTIUM Research Team
A systematic audit of coverage quality across a broad set of the world's languages, including many low-resource languages, across eight state-of-the-art multimodal models. We introduce coverage indices for morphological handling, code-switching robustness, and script fidelity, and propose training interventions that measurably narrow the coverage gap.
Collaboration
Research is a conversation.
We collaborate with academic institutions, government research bodies, and independent researchers around the world. If you are working on problems that intersect with our research areas, such as federated architectures, multilingual NLP, legal AI, or safety, we want to hear from you.
We offer research access to our models for academic and institutional researchers. We also participate in joint publication and co-supervision of doctoral research. The problems we are working on are too important for any single team to solve alone.
Work With Us
Join the research effort.
We are recruiting researchers at all levels, from doctoral interns to principal scientists. We are also open to institutional collaborations and research partnerships.