# [[Domain-Specific SLMs for Risk Intelligence]] Small language models fine-tuned on digital infrastructure risk data. Not general-purpose chatbots. Specialized models that understand power purchase agreement structures, environmental permitting language, grid interconnection dependencies, and data center site selection criteria. Training corpus: regulatory filings, environmental impact assessments, grid reliability reports, insurance actuarial data, infrastructure deal documents, market analyses, industry research. Why SLMs, not LLMs: speed matters for real-time risk scoring. Cost matters at portfolio scale. Privacy matters with proprietary deal data. A compact model fine-tuned on infrastructure risk outperforms GPT-4 on domain-specific risk assessment tasks while running privately. The competitive advantage: these models encode domain expertise that currently lives inside a handful of specialist consultancies. When Azraq deploys an SLM trained on thousands of risk assessments, it democratizes that expertise across every deal on the platform. Combined with [[Azraq Data Agents]], these models continuously improve. New regulatory filings get ingested, new site data refines risk predictions, new deal outcomes calibrate accuracy. The knowledge compounds. Links: - [[Azraq MOC]] - [[Azraq Data Flywheel]] - [[Foundational Models MOC]] --- Tags: #deeptech #kp