Skip to main content

UC Data Science Symposium 2025 — Summary

Keynote 1: Ryan Fitzpatrick (GE Aerospace)​

Topic: The Evolution of Analytics & AI at GE Aerospace Summary: Ryan highlighted GE Aerospace’s global scale and how AI is being embedded across its value chain. From supply-chain optimization to predictive maintenance.

  • Applications: AI predicts part replacements, enhances fuel efficiency, and reduces supply bottlenecks.
  • Integration: AI assistants are being used internally for decision support and workflow automation.
  • Challenges Discussed: Maintaining data quality and consistency across datasets from both legacy and modern engine systems.

Keynote 2: Matt Booher (E.W. Scripps Company)​

Topic: Data Goes Live: How Scripps Is Evolving AI & Analytics Summary: Matt described how AI has transformed the media industry and the way Scripps leverages analytics for real-time storytelling and audience engagement.

  • Evolution: From spreadsheets → predictive modeling → AI-driven systems.
  • Transformation Goals: Integrating human judgment with AI-powered systems to enhance content delivery and business agility.
  • Focus: Building responsible, explainable AI systems that drive measurable impact while preserving editorial integrity.

Breakout Session 1: Matt Ritchey (GAIG)​

Topic: Forecasting Catastrophe Insurance Losses with AI — Integrating Weather Models, Computer Vision, and LLMs

Core Concepts:

  • Multi-modal AI framework combining:

    • Numerical Weather Prediction
    • Computer Vision (spatial feature extraction)
    • Large Language Models (semantic parsing of hurricane advisories)

  • Objective: Predict total insured losses from catastrophic weather events 24–36 hours before landfall.

  • Data: Historical and open-source weather data (2016–2024).

  • Findings:

    • Adding LLM-based semantic features improves accuracy and interpretability.
    • Highlights adaptability to other disaster types (e.g., wildfires).

  • Discussion:

    • Currently hosted on Azure GPU.
    • Does not yet incorporate feedback correlation between predictions and real-world outcomes for retraining.

  • Reading Recommendations:

    • Practical Deep Learning for Coders (fast.ai)
    • Deep Learning with Python (François Chollet)

Breakout Session 2: Dave Ramsey (Syrv.AI)​

Topic: Implementing ReACT Agents from Scratch

Overview: Dave explained how to construct ReACT-style (Reason + Act) AI agents manually without using pre-built frameworks.

  • Core Topics:

    • Action Planning: Structuring multi-step reasoning and decision flows.
    • Memory Management: Persisting and recalling contextual information effectively.
    • Feedback Loops: Designing iterative reasoning cycles for agent refinement.
    • Tool Integration: Managing how agents interact with APIs or knowledge bases.

  • Book Reference: Claude Code: The Complete Guide (authored by Dave Ramsey).

  • Takeaway: Understanding the internal architecture of autonomous agents helps in customizing their reasoning, responsiveness, and efficiency beyond typical framework constraints.

Overall Reflection​

The symposium showcased how AI and analytics are moving from predictive insights to autonomous, real-time decision systems in aerospace, media, insurance, and AI agent design. Each session emphasized the growing importance of data integrity, interpretability, and responsible AI integration in complex, real-world operations.

Last updated on by Rohit Kulkarni