Where others see systems
we see structures.
AENEA Global builds open-source AI infrastructure and applies it to systematic intelligence for institutions and governments. Our latest releases — BHR-0.1, a 74.8M System 1 decision engine with 98% contradiction detection, and Pinta-2.0, a 74M semantic router with 92.45% accuracy — ship with mathematically calibrated probabilities and edge-native deployment. We control the data, the architecture, and the applied endpoints.
AENEA Quartz
Our primary division — and the foundation everything else is built on. AENEA Quartz unifies our data infrastructure (Quartz), our tokenizers (QT-Cittern-12k), and our language model laboratory (AENEA) into a single vertically-integrated AI operation. Two engines, one architecture: Pinta routes traffic, BHR makes decisions. Apache 2.0 licensed, open-source by default.
Built on the novel CartanLM architecture with Block Householder Linear layers constraining weights to the Stiefel manifold. Orthogonal geometry preserves semantic boundaries that dense models collapse. Beats 421M dense models on specialist tasks. Ships fully containerized in Docker.
Read the BHR-0.1 model card →Built on the 24-layer Cittern-2.0 architecture with block Householder orthogonal attention. One forward pass classifies prompts into nine reserved domain tokens for dynamic MoE dispatch. Ships with ONNX and a zero-dependency C++ edge router.
Read the Pinta-2.0 model card →Crassus
Credit and financial research website with eleven published reports on distressed credit, structured products, and systemic risk. Coverage spans consumer lending, utilities, insurance, and SPAC markets. All reports available at crassus.info.
We build the AI. We apply it to financial and government systems.
By owning the data pipeline (Quartz) and the tokenizer layer (QT-Cittern-12k), we produce language models (AENEA) that achieve factual recall at a fraction of the computational cost. Our two flagship engines — BHR-0.1, a 74.8M decision engine with near-perfect probability calibration, and Pinta-2.0, a 74M semantic router — both run on the orthogonal Block Householder architecture, proving that geometric preservation scales down better than dense parameter scaling.
This deterministic accuracy powers our applied research published via Crassus. The building is the capability. The application is the value. From open-source models and tokenizers to published credit research, our vertically integrated AI infrastructure delivers analytical depth without legacy infrastructure cost.
Published Analysis via Crassus
Institutional-grade credit research, live at crassus.info
The Sports Sponsorship Signal
Non-financial leading indicator of credit sector stress. 100% historical hit rate 2004–2008. Original research.
The Consumer Credit Ecosystem
Mapping the $200bn pipeline from fintech origination to pension fund absorption. Five firms, all originators.
Thames Water: Liquidity-Gated Restructuring
Regulatory-constrained distressed credit. AGO ~$1.04bn expected loss.
BNPL Sector: Systemic Risk Profile
Funding is procyclical. Warehouse + ABS tighten exactly when losses rise.