>> Deep Thinking & Reasoning Machines <<
We are focused on AI systems that creatively solve the most challenging analytical problems. We combine large and small language models, test-time heuristics, supervised fine tuning, reinforcement learning, code generation and agentic patterns optimized for reasoning, idea exploration and discovery. We use graphs and network analysis, custom novel ontologies, semantic similarity, and code-mediated reasoning and evaluation.
We believe that next level AI -and possibly AGI- will emerge from multi-entity orquestration, with a mix of high level language models and lower level, specialized models that operate in more abstract spaces; plus memory, orquestration agents, among other strategies.
> [potential] target domains (current SOTA is brittle or impossible)
* Exact global optimization under uncertainty * Long-horizon dynamic programming & combinatorial search * De novo protein design & inverse folding * Quantum chemistry & materials discovery * High-dimensional resource allocation with logical constraints * Symbolic regression over massive experimental datasets
> Advantages are structural, not incremental
* novel, creative solutions via deep exploration of the solutions space * 10–100× lower total cost than human experts alone * unparallel privacy potential * modularity * Full audit trails with code and mathematical certificates * Orders-of-magnitude higher precision on constrained problems
> Current Status
building in stealth mode. express interest.
▓