AI & ML interests
Small Language Models, Efficient Training, Efficient Inference, Model Architecture, Reinforcement Learning, Agents, Multimodal AI, Long-Context & Memory, Open-Weight Models
Recent Activity
EXOFORCE AI
Building efficient intelligence without datacenter-scale hardware.
Language Models · Training · Inference · Agents · Memory · Reinforcement Learning
What we build
EXOFORCE AI is the AI research and engineering division of EXOFORCE.
We build and experiment with efficient language models, training systems, inference infrastructure, agents, and memory architectures designed around real hardware constraints.
Research
- Small Language Models — capability at compact parameter counts
- Efficient Training — getting more from limited compute and VRAM
- Inference — latency, throughput, memory usage, and deployment
- Model Architecture — experimenting beyond default Transformer recipes
- Agents — models interacting with tools, environments, and other systems
- Memory — long-term and external memory beyond the context window
- Reinforcement Learning — learning through interaction and feedback
- Multimodality — extending models beyond text
Models
Antares
Experimental language models focused on efficient architectures, constrained-hardware training, and small-scale research.
Zeon
Language models used for pretraining, fine-tuning, post-training, and capability experiments across EXOFORCE AI.