News

We've released version 2.0 of our foundational paper: "Neuro-Semiotic Reasoning: Algebraic Semiosis for Observational Causal Inference."

What's new: This version reports the key theoretical finding that recursive semiotic interpretation — predicted by Peircean theory to require iterative processing — admits algebraic collapse. The recursive chain can be computed as constant-time resonance operations, achieving equivalent or superior reasoning quality at 140× lower latency than LLM-based alternatives.

We validate this empirically through a systematic ablation study across 14 experimental configurations and 109 evaluation queries. The algebraic architecture achieves a mean answer quality of 4.4/5.0 on causal reasoning tasks — outperforming the LLM baseline (4.1) while producing zero context-exhaustion errors. Adversarial falsification proved the most robust capability, scoring a perfect 5.0/5.0 on 14 of 24 adversarial tests.

A counterintuitive finding: stripping raw evidence text from the pipeline and providing only structural facts improves causal narration quality (4.6/5.0). This supports the core thesis — reasoning substrates should be structural, not textual.

Download the full paper (v2.0) · Research brief (2-page summary)

We've released version 1.5 of our foundational paper, "Neuro-Semiotic Reasoning: Computational Semiosis for Observational Causal Inference."

Key Update: This version introduces Semantic Atomicity as a fundamental computational constraint. While traditional generative models and NLP extractors typically produce "fuzzy," compound relationships that mirror the ambiguity of human language, the Neuro-Semiotic architecture now enforces a strict decomposition of all observational signals into indivisible $(s, p, o, t)$ tuples (subject, predicate, object, timestamp).

By enforcing semantic atomicity at the point of ingestion, we ensure that the subsequent hyperdimensional binding and causal simulation operations are mathematically pure, reversible, and fully auditable. This bridges the gap between linguistic nuance and the formal rigor required for high-stakes causal discovery.

Download the updated paper (v1.5).

We're releasing our second research paper: "The Abductive Hypothesis Engine: Evolving Observational Causal Inference."

The core idea: LLMs shouldn't reason at runtime — they should write the reasoning algorithms offline. This paper presents an architecture that demotes the LLM from a runtime reasoner to a linguistic renderer, shifting causal inference onto deterministic, evolved algorithms running on a mathematically verifiable substrate.

The system adapts evolutionary coding (in the lineage of AlphaEvolve) to real-world observational data — evolving graph-traversal algorithms that explain structural absences in temporal knowledge graphs. When causal chains are broken by missing data, the engine performs abductive counterfactual simulation to infer hidden links, grounded in the invariant-preserving algebra of Hyperdimensional Computing.

This builds directly on our Neuro-Semiotic Reasoning framework and represents the theoretical foundation for o-machine's next-generation reasoning capabilities.

The paper will be available on SSRN and arXiv shortly. Follow Martin on LinkedIn for the release announcement.

We've published version 1.3 of our foundational paper, "Neuro-Semiotic Reasoning: Computational Semiosis for Observational Causal Inference."

The core contribution: Introducing Behavioral Signals as a different unit of computation. LLMs see tokens—text fragments stripped of time and context. o-machine sees behavioral signals—hiring, partnering, facility expansions, absent events—anchored in time and modality-independent (extracted from text, satellite imagery, sensors, transactions).

This isn't an incremental improvement to existing architectures. It's a different foundation for reasoning about what's happening in reality before it becomes a document.

Download the paper or read more about our approach on the research blog.

We're opening signups for our closed beta, launching in Q2 2026.

We're inviting 20 early users to help shape o-machine. If you're a private market investor or autonomous driving industry professional who needs to understand causation—not just correlation—we'd love to hear from you.

Request beta access and share your use case. We'll be in touch if it's a good fit.

We're heading to Mountain View for NVIDIA GTC and staying through the entire month of March to meet investors and teams building at the frontier of AI reasoning.

What we're building: o-machine shows you how things actually connect — and proves it. We discover hidden actors and relationships, trace causal chains, and explain why things happen. Every connection backed by evidence.

Our current focus is private market intelligence: know about private companies before anyone else. We reason about what didn't happen (absence detection), track velocity of change, and discover hidden relationships that statistical models can't see.

The core difference: Statistical AI optimizes for correlation. o-machine optimizes for causation. Different architecture. Different capabilities. Different results.

Our paper on the underlying neuro-semiotic architecture will be available on Arxiv in early March.

Connect with us: If you're working on reasoning infrastructure, private markets intelligence, supply chain, or defense — or investing in teams that are — we'd love to talk. Reach out to our co-founder Martin on LinkedIn or email hello@o-machine.com.

We're submitting our first research paper to Arxiv: "Neuro-Semiotic Reasoning: Computational Semiosis for Observational Causal Inference."

The paper formalizes the architecture behind o-machine: how causal meaning emerges through recursive interpretation of differential relationships across temporal, spatial, conceptual, and absence dimensions.

The core insight: Statistical AI optimizes for correlation (what appears together). o-machine optimizes for causation (what relationships mean). This enables capabilities structurally impossible in LLMs: reasoning about what didn't happen, tracking velocity of change, discovering entities through behavioral patterns, and building complete audit trails for every inference.

This is the first of several papers we're releasing this year. We're building in public and sharing our research as we go.

The paper will be available on Arxiv in early March. Follow Martin on LinkedIn for updates.

We're heading to Mountain View for NVIDIA GTC and staying through the entire month of March to meet investors and teams building at the frontier of AI reasoning.

What we're building: o-machine shows you how things actually connect — and proves it. We discover hidden actors and relationships, trace causal chains, and explain why things happen. Every connection backed by evidence.

Our current focus is private market intelligence: know about private companies before anyone else. We reason about what didn't happen (absence detection), track velocity of change, and discover hidden relationships that statistical models can't see.

The core difference: Statistical AI optimizes for correlation. o-machine optimizes for causation. Different architecture. Different capabilities. Different results.

Our paper on the underlying neuro-semiotic architecture will be available on Arxiv in early March.

Connect with us: If you're working on reasoning infrastructure, private markets intelligence, supply chain, or defense — or investing in teams that are — we'd love to talk. Reach out to our co-founder Martin on LinkedIn or email hello@o-machine.com.

We're excited to announce that o-machine will be attending the World Economic Forum in Davos, Switzerland for Promenade events from January 19-23, 2026.

If you're attending and would like to connect, reach out to our co-founders Irina or Martin on LinkedIn.