Agentic Fundamentals: How language models and code can communicate
summary
Claims developers can reliably make code talk to LLMs and pre-trained models to turn messy documents into usable data, proven on commercial leases (often 50–80 pages) and emails. Uses OpenAI API/SDKs, InstructorJS, LangChain, and schema tools like Zod to send a JSON schema and get structured JSON back. Notes 80–90% of data is unstructured, explains delimiters and context-window limits, and shows the extractor running in real client workflows.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:26AI advocacy and real-world use
- 3:56AI chef versus AI architect
- 4:59Commercial lease automation
- 8:09History of language models
- 11:03Context in language models
- 12:27LLMs as interfaces
- 14:29Unstructured data challenges
- 15:52Structured output for LLMs
- 18:19Parsing emails with LLMs
- 20:45JSON schemas for output
- 23:12Contact info extraction code
- 24:36Context engineering and prompts
- 25:18Summary
related talks
47mStructured Outputs in OpenAI: Building Reliable JSON-Based AI Integrations
01h 59mLLM applications with LangChain
32mDocuments Meet LLMs: Tales from the Trenches
47mJava Meets AI: Integrating LLMs into Your Applications with LangChain4j
01h 34mA Practical Starter's Guide to building LLM based projects
26mLanguage Model Agents Enable Semi-Formal Programming
34mProgramming Languages for AI Programing Agents
24mHow Generative AI Apps Use APIs