Structured Outputs in OpenAI: Building Reliable JSON-Based AI Integrations
summary
Shows production patterns that force LLMs to emit JSON via the OpenAI responses API using developer-defined JSON schema so outputs are type-safe and reduce hallucination. Demonstrations extract contract fields (title, effective date, duration, parties, enum license scope) from PDFs, evaluate parsers (Mistral, Dockling, Llama, OpenAI) for a RAG pipeline, and run a chat agent that uses function calling with strict true and zod/pydantic schemas (no optional fields, additionalProperties false, co‑speaker max 3). All demos run end‑to‑end; code and slides sit in the linked GitHub.
AI-written summary · about
chapters (7)
- 0:00Structured outputs in AI
- 1:12What structured output is
- 10:01Structured output with OpenAI API
- 18:30Live demo: PDF data extraction
- 30:02Function calling with structured output
- 38:27Live demo: talk submission agent
- 45:29Resources and closing
related talks
17m"Taming LLMs with Structured Outputs"
26mAgentic Fundamentals: How language models and code can communicate
02h 08mIntroduction to AI Agents
01h 00mComplex Data Ingestion with Open Source AI
49mNext-Gen Documentation with AI
34mJavaScript-Powered AI: Crafting Intelligent Systems with LLMs and Vector Databases
33mOperationalizing GenAI Applications with DevOps Practices
01h 13mBuilding Production-Ready AI Systems in 90 Minutes