Greg Michaelson

Moving beyond complex LLM-as-a-judge metrics: Opik's Test Suites offer a pragmatic way to simplify AI evaluation by using plain-English assertions to validate agent behavior instead of just interpreting wall-of-text decimal scores. https://lnkd.in/e9PE2Pwt #MachineLearning #AI #LLM #DeepLearning #AgenticAI

1

In this world of Agents.. how do you ensure that your agent is doing what it's designed to do? Agent Evaluation is one of the core capabilities required for Explainable AI. @Snowflake #Cortex #Agents Evaluation is now GA.

19

The right tools in the hands of a craftsman can result in amazing things. In neuroscience, our "tools" are the architectures that handle complex, multimodal brain data. When we treat data engineering as a craft - focusing on precision and scalability - we move from storing bits to enabling breakthroughs.

2

Of Note: "A new hybrid neural network framework inspired by biological systems for advanced financial forecasting" Published: October 22, 2025 Lately, I've gone back to my data origins, hyperfocusing on methods for analysis and evaluation for the lines of business I always thought simple to predict: finance and banking. The techniques and tools available made it "too easy" to forecast market responses and anticipate trends and I recognized rather quickly how data driven insights transformed corporate responses to market dynamics and consumer needs. It represented the exact opposite end of the spectrum that led me to Data Science in the first place. However, recently I challenged myself to look at it a bit differently. If I am as good as I know I am, then it is imperative and of vital importance for me to create better fintech product and solutions. The reason being, ideating, designing and building models to fix problems I have zero connection to for the exact type businesses I told myself I'd never promote nor endorse will allow me masterfully create more innovative and essential algorithms overall. Also, working hard creating something you would rather not, increases your cognitive reserves and brain health.

5

The real shift isn’t from semantic layer to ontology. It’s from description to execution. A recent and excellent Metadata Weekly article by Jessica Talisman explores the shift from traditional semantic layers to ontologies and context graphs for AI. It makes a valid point: Many semantic layers were built for dashboards. They define metrics and standardize business logic, but primarily for human reporting. AI systems require something deeper. They need structured meaning: entities, relationships, constraints, inheritance, and context. In short, they need ontology. But there’s an important nuance in this evolution. Traditional ontologies (such as OWL-based models) are excellent at describing and reasoning over domain knowledge. Yet they weren’t designed to operationalize enterprise metrics at warehouse scale. If metrics are to power AI, not just dashboards, they must be more than definitions. They must be executable. That’s where SQL ontologies introduce a meaningful shift: • Relationships are modeled as first-class constructs • Metrics are governed and reusable • Measures execute natively in the data warehouse • AI consumes executable business logic, not just metadata The conversation is no longer “semantic layer vs ontology.” It’s about building a semantic foundation where meaning and execution coexist. As AI becomes embedded in enterprise workflows, treating metrics as first-class, SQL-native semantic objects may be the real differentiator. #Ontology #SemanticLayer #EnterpriseAI #KnowledgeGraph #DataArchitecture #GenAI

152

34 Comments

Team Enqurious is proud to partner with Databricks Community team to amplify their Databricks upskilling initiatives. Building is the new Learning. So, why not starting dirtying hands from the get go? Stay tuned for more! Prateek Kumar Sayli Nikumbh Mansi Mutreja Shwetabh Varma Amit Murarka #BrickbyBrick #Lakehouse #Genie #GenAI

15