Raleigh AI, Machine Learning, and Computer Vision Meetup

In Through the Looking-Glass, Lewis Carroll envisioned a world where logic is inverted, rules shift unpredictably, and progress requires very strategic navigation. And if you have attempted to use an open source large language model within an enterprise setting then you have had first hand experience with unpredictable behavior. Today, one of the possible mitigation strategies to address this unpredictability is known AI agents or agentic. Software applications (or bots) that operate in a similarly dynamic landscape—processing vast amounts of unstructured data, collaborating in multi-agent workflows, and self-correcting in ever-changing environments.

High-quality models require high-quality data, yet identifying problematic samples in massive datasets remains challenging. This session introduces a robust framework leveraging Databricks, FiftyOne’s Data Lens, and Mosaic AI to enhance data quality and model performance. Data Lens enables intuitive querying of billion-scale data samples in a data lake, while Mosaic AI uncovers hidden relationships through multimodal embeddings. Combined with Databricks’ scalable data management, these tools streamline error detection, model refinement, and dataset optimization, advancing the development of reliable foundation models in Visual AI.