From Click to Conversation: Let's Replace Buttons and Forms with an LLM!
summary
A prototype replaces vehicle registration screens with a chatbot that accepts create/update/delete operations for model, year, and owner; it’s implemented in Python with a Gradio demo and uses the OpenAI API. The talk compares three engineering approaches—one huge prompt, a state-machine built from prompt fragments, and LangChain agents that call real Python tools—and demonstrates live runs showing agents need GPT-4 for reliable behavior while GPT-3.5 often fails, with discussion of cost, prompt brittleness, prompt injection, retrying JSON outputs, and GDPR risk.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 7:07Problem and first solution
- 13:10Simple solution demo
- 18:21State machine approach
- 23:27Agent and tools solution
- 29:19Agent demo
- 31:02Production issues
- 39:53Conclusion
related talks
51mChatGPT as a developer's personal servant
01h 00mGoing beyond ChatGPT: introduction to prompt engineering & LLMs
47mAI-Powered Form Wizards: Chat, Click, Done
01h 39mMastering LLMs: From Prompt Engineering to Agentic AI
01h 04mBuilding Expertise Into Generated Conversations
58mA Primer on AI Powered Conversational Experiences and Language Models
45mWeb Development in the Age of AI
55mEveryday AI: Practical Applications Beyond the Chatbot