Thomas in't Veld | AI Track | Compass AI & Data Summit 2025
summary
Claims that agentic analytics can turn natural‑language questions into repeatable, testable SQL and deliver actionable marketing insights rather than dashboards, using LLMs plus structured data and a semantic layer. Demo contrasts an inconsistent oneshot OpenAI prompt with a reproducible pipeline built with Dispy (declarative self‑improving Python), schema and join rules, synthetic data in DuckDB, Faker and observability; the agented triage→SQL flow runs reliably and surfaces falsifiable, hypothesis‑driven recommendations.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 5:02AI adoption challenges
- 8:09Agentic systems hype
- 11:58Marketing insights & analytics
- 15:49Modern data stack placement
- 20:07Three rules for agentic deployments
- 23:22Engineering with Dispy
- 28:13Agentic analytics demo
- 33:23Summary
- 35:30Q&A on AI impact
- 37:35Q&A on ROI studies
- 39:40Q&A on implementation
- 42:57Q&A on AI future
- 44:19Q&A on agentic systems
related talks
45mWhat’s Next in AI for Data and Data Management
19mSession: Game Data For All: How AI Is Democratizing Game Analytics
23mTrusted Agentic AI:Architecting Systems with Reliable Data at Scale
17m"Data readiness" is a Myth: Reliable AI with an Agentic Semantic Layer
32mSponsored Session: Observability and Agentic AI - From Observability to Answers...
26mAI & SRE: Embracing the Ultimate Frontier of Automation
29mLeveraging AI to Enhance DevEx & Operational Efficiency
16mGame Data for All: How AI Is Reinventing Game Analytics