Naveen Rao, Databricks ｜ RAISE Summit 2025

Context & Setting

• The discussion took place at the RAID Summit in Paris, France, a two-day international conference focused on the future of AI.
• Key industry themes identified include Sovereign Cloud, data platform evolution, and infrastructure, with significant attention paid to the relationship between data and intellectual property.

Strategic Pivot in Data Utility

• The industry is shifting from treating data as a raw asset to viewing it as the essential "ingredients" required to create valuable, differentiated AI assets.
• Analytics is characterized as "Step One" for business understanding, whereas current strategy focuses on using data to define IP and secure competitive market advantages.
• Databricks' thesis, established through the Mosaic ML acquisition, centers on customizing AI models to unlock value from processed data.

Evolution of Data Application

• The application of data is morphing from simple model training to:
  • Contextualizing AI responses.
  • Fine-tuning models for specific use cases.
  • Building evaluation criteria to assess emerging model types.

The Critical Role of Evaluation

• Evaluation is defined as the process of explicitly defining success goals and grading outcomes on a 0–100% scale before model execution.
• This approach addresses the 2023 trend of "magic thinking," where organizations expected value generation merely by feeding data into models without clear success criteria.
• Jonathan Frankel, former chief scientist, advocates for automating this evaluation process to allow for customization and continuous checking of model performance.
• The evaluation process relies heavily on synthetic data and agent-driven methods supported by probability theory.

Methodological Inversion

• Traditional machine learning defined the model by the data (e.g., regression lines fitting data points).
• The current paradigm inverts this by hypothesizing a model of the world first and using observational data to validate or falsify that hypothesis.

Platform Convergence & Feature Rollouts

• A convergence is occurring between platform/data engineering and business logic, disrupting traditional dashboarding silos.
• Databricks has integrated a dynamic feature called "Genie" into its dashboarding tools.
  • This allows users to query data directly via natural language rather than relying on static graphs.
  • The system automatically generates SQL code, processes the data, and plots the results on demand.

Forward-Looking Trajectory

• The immediate progression for IT and database professionals involves moving from manual SQL management to utilizing automated, dynamic data querying systems.
• The intersection of analytics and Gen AI is identified as a key disruption point where business logic becomes deeply embedded in the data platform layer.