Est. 2019 —classified research division—

NEXUS
AI RESEARCH
LABORATORY

compute_nodes : 2,847 active
neural_nets : 1,204 training
throughput : 847.3 PFLOPS
anomalies : 3 detected
Active Experiments LIVE
NLP Training Runs 347
CV Model Evaluations 128
RL Agents Active 89
Cluster Health NOMINAL
GPU Utilization 94.7%
Memory Allocation 87.2%
Network I/O 42.8 GB/s
Scroll to explore
// Research Domains

Pushing the Frontier
of Machine Intelligence

6 active research verticals
142 researchers deployed
38 publications this quarter
01
🧠
Large Language Models
& Foundation Architectures
Developing next-generation transformer architectures with emergent reasoning capabilities. Our NEXUS-7B model achieves state-of-the-art performance on mathematical reasoning benchmarks while maintaining efficient inference through novel sparse attention mechanisms.
Transformers Sparse Attention RLHF Chain-of-Thought Reasoning
02
👁️
Computer Vision
& Multimodal Perception
Unified vision-language models enabling machines to perceive, understand, and reason about visual content with human-level accuracy across 47 benchmark categories.
Vision Transformers CLIP Segmentation
03
⚡
Reinforcement Learning
Multi-agent systems and reward-shaping techniques for complex strategic decision-making.
Multi-Agent PPO
04
🧬
AI for Science
Accelerating drug discovery and protein structure prediction through geometric deep learning.
Protein Folding Molecules
05
🔊
Audio &
Speech
Intelligence
Real-time transcription, voice synthesis, and acoustic scene analysis.
Whisper TTS
06
🔒
AI Safety & Alignment
Interpretability research, red-teaming methodologies, and constitutional AI frameworks for trustworthy systems.
Interpretability Red-Team
07
🔬
Efficient Inference
Quantization, distillation, and edge deployment strategies for production-grade AI systems.
Quantization Edge AI
08
🌐
Robotics & Embodied AI
Bridging perception and action in physical environments through sim-to-real transfer learning.
Sim2Real Manipulation
// Live Training Monitor

Real-Time Model Training

Model: NEXUS-13B-v4
Dataset: RedPajama-v2 (1.2T tokens)
Status: Training epoch 47/100
Loss Curve — Cross-Entropy
Train Loss
0.847
Val Loss
0.912
Epoch
47
Neural Architecture
Model Specifications
NEXUS-13B-v4
Sparse Mixture-of-Experts transformer with 13B total parameters, 2B active per forward pass.
Parameters
13.2B
Layers
40
Attention Heads
64
Context
32K
// Research Output

Selected Publications

384 total publications
12,847 aggregate citations
h-index: 94
2024
NeurIPS 2024 Spotlight
Sparse Mixture-of-Experts with Dynamic Routing for Efficient Multi-Task Learning
Chen, W., Nakamura, K., Osei, A., Petrov, D., Liu, M. & Rivera, S.
0
Citations
2024
ICML 2024
Constitutional AI: Scalable Oversight through Recursive Self-Improvement
Rivera, S., Chen, W., Müller, T., Gupta, P. & Johansson, E.
0
Citations
2024
Nature ML
Geometric Deep Learning for Protein-Ligand Binding Affinity Prediction at Atomic Resolution
Müller, T., Okonkwo, C., Nakamura, K. & Chen, W.
0
Citations
2023
NeurIPS 2023 Oral
Emergent Reasoning in Language Models via Chain-of-Thought Distillation
Liu, M., Gupta, P., Osei, A., Rivera, S. & Johansson, E.
0
Citations
2023
arXiv 2023
Neural Architecture Search for Vision Transformers with Hardware-Aware Constraints
Petrov, D., Nakamura, K. & Chen, W.
0
Citations
2023
ICML 2023
Multi-Agent Reinforcement Learning with Communication Compression at Scale
Johansson, E., Osei, A., Gupta, P. & Müller, T.
0
Citations
// Personnel Roster

Research Scientists

142 researchers worldwide
18 countries represented
94% hold PhD degrees
Dr. Wei Chen
Chief Scientist / Director of Research
ID: NX-001 · Clearance: ALPHA
h-index: 67 · Papers: 184
Focus: Foundation Models, NLP
Transformers LLMs RLHF Reasoning
Dr. Sofia Rivera
AI Safety Lead
ID: NX-007 · h-index: 42
Focus: Alignment, Interpretability
Constitutional AI Red-Teaming
Dr. Kenji Nakamura
Vision Intelligence Lead
ID: NX-012 · h-index: 51
Focus: Multimodal, Vision
ViT Segmentation
Dr. Priya Gupta
RL Systems Architect
ID: NX-023 · h-index: 38
Focus: Multi-Agent, Robotics
PPO Sim2Real
Dr. Adaeze Osei
Audio & Speech Lead
ID: NX-031 · h-index: 29
Focus: Speech, Acoustics
Whisper TTS
Dr. Thomas Müller
AI for Science Lead
ID: NX-018 · h-index: 45
Focus: Drug Discovery, Proteins
GNN Molecules
Dr. Mei Liu
Reasoning Systems
ID: NX-044 · h-index: 35
Focus: Chain-of-Thought, Logic
CoT Distillation
// Compute Infrastructure

847 PFLOPS
of Raw Compute

Our distributed training cluster spans three data centers across two continents. Custom-built GPU nodes interconnected with 400GbE fabric enable training runs that would take months on commodity hardware to complete in days.

LOCATION: [CLASSIFIED] · Geneva, CH & Reykjavik, IS & Singapore, SG
NETWORK: 400GbE InfiniBand fabric
COOLING: Immersion liquid cooling (PUE 1.04)
POWER: 47MW (100% renewable)
0
PFLOPS
Peak Compute
0
GPU Nodes
0
PB
Storage
99.97%
Uptime SLA
47MW
Power Draw
// Personnel Acquisition

Join the Mission

We recruit researchers who see problems others don't. If you're drawn to the unexplored edges of machine intelligence, we want to hear from you.

Open Positions
All roles include relocation support, unlimited compute budget, and equity participation.
Research Scientist — Foundation Models
Geneva · Remote Ok
ML Engineer — Training Infrastructure
Reykjavik · On-site
Alignment Researcher
Singapore · Remote Ok
Computer Vision Scientist
Geneva · Hybrid
RL Engineer — Robotics
Reykjavik · On-site