Stenobird

Episode

Orion at Gravity: Trustworthy AI Analysts for the Enterprise

Podcast
Data Engineering Podcast
Published
Mar 8, 2026
Duration seconds
3901
Processing state
processed
Canonical source
https://www.dataengineeringpodcast.com/gravity-orion-agentic-analytics-for-enterprise-episode-504
Audio
https://op3.dev/e/dts.podtrac.com/redirect.mp3/serve.podhome.fm/episode/f6ff0caa-931b-4c08-bfdd-08dc7f5cd336/6390860976111881687bb6b052-3f3e-42a6-93d8-f50f5538cc8c.mp3
JSON
/v1/public/podcasts/data-engineering-podcast/episodes/orion-at-gravity-trustworthy-ai-analysts-for-the-enterprise
Markdown
/podcast/data-engineering-podcast/orion-at-gravity-trustworthy-ai-analysts-for-the-enterprise.md

Actions

  • POST https://stenobird.com/v1/public/podcasts/data-engineering-podcast/episodes/orion-at-gravity-trustworthy-ai-analysts-for-the-enterprise/transcription-requests
    Idempotently request low-priority transcript generation for this episode.
  • GET https://stenobird.com/podcast/data-engineering-podcast/orion-at-gravity-trustworthy-ai-analysts-for-the-enterprise.md
    Read the agent-friendly Markdown representation of this episode resource.

Summary

The founders of Gravity discuss moving beyond static dashboards toward 'agentic analytics' using semantic layers and business context. They explain how Orion acts as a virtual coworker by combining structured data governance with real-time business intelligence.

Topics

  • Agentic Analytics
  • Semantic Layer
  • Context Engineering
  • Data Governance
  • Generative AI
  • Enterprise Data Strategy
  • Business Intelligence
  • LLM Observability

Highlights

  • Main idea: AI analytics must shift from one-shot text-to-SQL queries to multi-turn dialogues that incorporate business context
  • Practical takeaway: Use semantic layers to provide the 'glass box' transparency needed for users to trust AI-generated insights
  • Failure mode: Relying solely on model capabilities without mapping complex, 'spaghetti' data structures leads to unreliable outputs
  • Main idea: Effective AI agents require 'context engineering'—integrating external signals like calendars and documents alongside database schemas
  • Practical takeaway: Focus on driving business actions and decisions rather than just delivering more metrics or dashboards

Chapters

  1. 1:00 Introduction to Context Engineering: Lucas Thelosen and Drew Gilson introduce the concept of applying semantic layers to agentic analytics.
  2. 5:50 Beyond Text-to-SQL: The shift from simple query generation to sophisticated AI analyst patterns that uncover the 'why' behind data.
  3. 10:30 Human-in-the-loop Governance: How data leaders can validate and correct AI findings to maintain control over business logic.
  4. 15:20 The Challenge of Legacy Data: Addressing the difficulty of building semantic layers over complex, fragmented data environments resulting from acquisitions.
  5. 20:30 The Evolution of Embeddings: How rapid advancements in LLM capabilities have changed the importance of traditional retrieval-augmented generation techniques.
  6. 25:10 The Glass Box Approach: Ensuring AI transparency by allowing users to trace insights back through transformations and Python scripts.
  7. 30:10 Modeling the Workday: Using external context, such as calendar events, to guide AI agents toward relevant, timely analysis.