Workshop: Getting Data Ready for AI - With Unified GraphQL &SQL Endpoints
summary
Comparison shows a GraphQL-only approach hands hundreds of rows into an LLM and produces hallucinated aggregates, while a new system called prompt ql gives LLMs a programmatic Python runtime plus SQL over a unified supergraph and deterministically translates SQL to underlying sources. Demo connects Postgres and Zendesk, runs an AI server with an Anthropic API key, computes top-five customers by summing invoices correctly, stores results as a structured memory artifact, calls downstream LLMs for summarization/classification, and gates mutations for user approval — working end to end.
AI-written summary · about
chapters (12)
- 0:00Workshop introduction
- 1:06GraphQL limitations for LLMs
- 2:34Setting up GraphQL API
- 8:50Building AI backend
- 13:25Introducing PromptQL
- 22:57Zendesk integration demo
- 29:57Enterprise use cases
- 36:44Taking actions with LLMs
- 40:00Privacy and self-hosting
- 43:38Logging and debugging
- 45:10Response quality benchmarking
- 51:16Q&A session
related talks
31mWhy AI needs GraphQL
15mTraining and Using Generative AI Models with GraphQL
26mEvolving GraphQL: Designing APIs That Deliver Insights with Intelligent Resolv…
14mKeynote: GraphQL in the AI Era
27mGraphQL Meets LLMs & Agents: Building Production AI at Starbucks Scale
45mForget SQL! Create your own AI Data Analyst
24mFrom Query to Conversation: GraphQL as an AI Interface Layer
29mKeynote: LLMs, what are they good for?