Est. 2026 · Jacob Westerbeck · Three Divisions
A proprietary scoring engine — the Positional Correlation Field — powers three distinct domains: commercial AI products, distributed agent infrastructure, and foundational protocol research. One method. Unlimited application.
Structure
Astrognosy AI operates across three distinct verticals, each built on the same proprietary PCF engine and protected by a growing patent portfolio.
Commercial AI Products
Production-ready AI infrastructure. Route LLM traffic intelligently to cut GPU costs from day one. Detect network intrusions without a training phase. Automate corpus management at scale.
pcfic.com →Agent Infrastructure
Distributed reasoning infrastructure for autonomous agents. WARF enables multi-agent arbitration, a shared corpus flywheel, identity-free personalization, and a global intelligence graph — at protocol level.
Live demo →Core Protocol & IP
The PCF engine, WARF RFC-001 specification, and a growing patent portfolio. Open protocol standards enable ecosystem growth while the core scoring method remains proprietary and defensible.
Read Whitepaper →Core Technology
The Positional Correlation Field (PCF) is a proprietary PMI-based scoring method — CPU-native, requiring zero labeled training data, and domain-agnostic by design.
// Positional Mutual Information — lag k
PMI_k(t_i, t_j) = log[ P(t_i,t_j,k) /
( P(t_i) · P(t_j) ) ]
// Decay kernel: K(k) = 1/(k+1), k = 1..K
PMI'_k = PMI_k × K(k)
// Normalized score per lag
σ_k = mean(PMI'_k) / (std(PMI'_k) + ε)
// Winner selection
S = aggregate(σ_1, ..., σ_K)
winner = argmax(S) // agent_id excluded
Live Demo
Three agents compete on the same query. The PCF engine selects the winner based purely on structural convergence with submitted evidence — agent identity excluded.
Loading…
Loading…
Loading…
Founder
"The best infrastructure disappears. It just works — faster, cheaper, and without retraining every time the world changes."
Astrognosy AI was founded by Jacob Westerbeck with a single conviction: that the most durable AI infrastructure is the kind that doesn't require labeled data, GPU compute, or a 30-day observation window to answer routine questions.
The PCF engine emerged from research into positional mutual information across large corpora — a method that proved equally effective for fact verification, network intrusion detection, and multi-agent reasoning arbitration. Three distinct domains. One proprietary engine. One growing patent portfolio protecting it all.