Let's Build a Supergraph Using ChatGPT
summary
Builds a Supergraph prototype by feeding ChatGPT prompts to generate GraphQL schemas, comparing GPT‑3.5 (fast, often messy) with GPT‑4 (slower, follows instructions better). Demonstrates concrete failures: missing pagination, no input types, wrong mutation payloads, and occasionally non‑existent federation directives. Shows fixes that work: adding a persistent system prompt, fine‑tuning a GPT‑3.5 model on GraphQL examples, and a custom UI that visualizes schemas and splits them into subgraphs; fine‑tuned model helps but needs much more training data to remove legacy mistakes.
AI-written summary · about
chapters (10)
- 0:00Introduction to supergraphs
- 3:54How ChatGPT works
- 8:46Demo with GPT-3.5 vs GPT-4
- 11:39Improving model responses
- 17:19Custom fine-tuned models
- 19:52Schema design limitations
- 20:33Future model improvements
- 22:41Visual schema editor demo
- 24:47System prompts vs fine-tuning
- 25:28How to try the tool
related talks
39mHow ChatGPT helps to boost your development efficiency
02h 00mChatGPT & Prompt Engineering
44mExploring ChatGPT for Improved Observability
01minteresting experience with ChatGPT
01h 01mHow ChatGPT Works
45mHow to use GPT in your projects: A Case Study
38mIntroducing the New Colleague - ChatGPT
15mTraining and Using Generative AI Models with GraphQL