Built on 15 Years
of Foundational Research

Where It Started

In 2008, Jake Westerbeck partnered with Jeff Spetalnick to build production NLP systems — practical tools for how machines handle language input, error correction, and statistical prediction from corpora. The research centered on a deceptively simple question: how does the structure of language — the patterns of which words follow which — encode meaning in a way that doesn't require brute-force compute?

That work produced two patents focused on frequency-based language efficiency: reducing keyboard input errors, improving OCR accuracy, enabling compression and translation through corpus-learned statistical likelihoods. These weren't academic exercises — they were built to run efficiently on the hardware of the time, without relying on large neural models that didn't yet exist at scale.

Jeff passed away before seeing where those ideas would lead. The research sat. But the fundamental insight — that positional co-occurrence patterns carry structural meaning that can be scored and compared without training — didn't go away. It waited for the right problem.

The Evolution to PCF

The leap from frequency-based NLP to Positional Correlation Fields was not a pivot — it was a completion. The 2010 patents worked because positional patterns in token sequences are structurally meaningful. PCF formalizes that intuition into a general-purpose scoring mechanism: analyze co-occurrence patterns across positional offsets, normalize by calibrated thresholds, and you have a structural signature — a vector that encodes what a sequence looks like, independent of domain.

When applied to network traffic flows in 2025, the same math that once predicted the next word in a sentence detected cyberattacks without a signature database. When applied to LLM query verification, it routed queries to CPU with 25% GPU savings. The engine was always domain-agnostic. It just took years of problems — and 15 years of perspective — to see it clearly.

Astrognosy AI was founded to build the products that prove this out: not as research demos, but as real infrastructure that enterprises can deploy today.

The Founder

Intellectual Property Timeline

From foundational frequency models to domain-agnostic positional fields — a 15-year arc of research, now protected across five product verticals.

Where This Goes

The Pacific platform exists to prove a single thesis: that a domain-agnostic structural scoring engine — one that runs on CPU, requires no training data, and improves automatically as it processes more queries — is the right foundation for AI infrastructure at scale.

Every product on the platform is a domain-specific proof of that thesis. Compass proves it in LLM routing. Laminar proves it in network security. Stratum, Forge, and Beacon are proving it in physical infrastructure. Wharf will prove it in multi-agent coordination — the port where agents offload cargo for PCF selection.