Publication

Model selection is a cost function, not a prestige contest. Define your quality floor, measure each candidate against it, then pick the cheapest model that clears the bar. Revisit quarterly. #DeepLearning #MLOps #AIInfra

If your agentic eval suite does not include adversarial tool inputs, it is not measuring real-world robustness. Agents that work perfectly in happy-path evals often collapse on malformed tool responses. #AIEngineering #AIAgents #AgenticAI