We are an AI research lab applying computational principles from cognitive and systems neuroscience to large language models. We work in the space between the agent layer and the model layer, on the inference-time computation that turns frozen weights and fixed prompts into useful reasoning.
Our thesis is that the next major gains in AI will come from organizing cognition dynamically at test time, under real-world constraints of accuracy, cost, and latency. We are building the systems and the science behind that thesis at the same time.
The team is small, technical, and unusually focused. Team members move quickly between research questions and production systems, and the loop between the two is short by design.
"The work demands research conviction and systems discipline: deriving algorithms from first principles, and shipping them as measurable infrastructure."
What the work asksTwo roles, one research program
We are currently hiring for two positions. Both report directly to the founder and contribute to the same research program, with different centers of gravity: one in systems, one in science.
Research Engineer, Test-Time Cognition for Agentic AI
We are looking for a strong research engineer to build the systems infrastructure behind our inference-time cognition layer for LLMs and agentic AI. You will implement algorithms for dynamic compute allocation, inference-time search, model orchestration, agent evaluation, and production-scale experimentation. This role is ideal for someone who can turn frontier research ideas into fast, reliable, measurable systems.
Research Scientist, Test-Time Cognition for Agentic AI
We are looking for a PhD-level research scientist to help define the scientific foundations of test-time cognition for LLMs and agentic AI systems. You will develop new algorithms for adaptive inference, uncertainty-aware compute allocation, search, verification, and multi-model cognitive orchestration. Our core thesis is that the next major gains in AI will come not only from larger models or more training, but from organizing cognition dynamically at test time under real-world resource constraints.
Send us a short note on what you would want to work on, along with your CV.
We read every application. A paragraph on a problem you have thought about, a paper you found surprising, or a system you have built tells us more than a long résumé. Mention the role in the subject line.
contact@voaige.com