Day 10

Context

Tasks (4)

1. Implement structured JSON output with schema (25 min)
   Call the Chat Completions API with Structured Outputs: response_format with type "json_schema" and a full JSON Schema definition. Extract structured data from unstructured text (e.g., extract name, role, company, email from a bio paragraph). Test with 5 different inputs. What is the failure rate? Compare to basic JSON mode. Save as /day-10/structured_json_extraction.js.
2. Build the API comparison table (25 min)
   Create a 3-column comparison: Chat Completions vs Responses API vs Assistants API v2. Rows: state management, context control, tool use, file handling, streaming, debugging transparency, cost attribution, web search, computer use. For each cell, write one sentence on the capability. This is production-useful reference material. Save as /day-10/api_comparison_table.md.
3. Test the Responses API (25 min)
   If you have API access: make a Responses API call with web search or file search enabled. Compare the developer experience to Chat Completions. If no access: read the documentation and write a 200-word assessment of when the Responses API is the right choice vs Chat Completions. Save as /day-10/responses_api_assessment.md.
4. API changelog review (25 min)
   Review OpenAI’s API changelog for the last 3 months. List the 3 most significant additions. For each: what use case does it unlock? Does it change architecture choices vs the Anthropic API? Save as /day-10/api_changelog_analysis.md.

Structured Outputs with JSON Schema — JavaScript

// OpenAI Structured Outputs — the recommended way to get reliable JSON
// This is now the DEFAULT for production extraction, not JSON mode

// Example: Extract structured contact info from unstructured text
const EXTRACTION_SCHEMA = {
  type: "json_schema",
  json_schema: {
    name: "contact_extraction",
    strict: true,
    schema: {
      type: "object",
      properties: {
        name: { type: "string", description: "Full name of the person" },
        role: { type: "string", description: "Job title or role" },
        company: { type: "string", description: "Company or organization name" },
        email: { type: ["string", "null"], description: "Email if mentioned, null otherwise" },
        confidence: {
          type: "string",
          enum: ["high", "medium", "low"],
          description: "Confidence in extraction accuracy"
        }
      },
      required: ["name", "role", "company", "email", "confidence"],
      additionalProperties: false
    }
  }
};

// Simulated API call structure
const apiCallStructure = {
  model: "gpt-4o",
  messages: [
    {
      role: "system",
      content: "Extract contact information from the provided text. Return structured JSON."
    },
    {
      role: "user",
      content: "Met Sarah Chen at the AI summit — she runs product at Anthropic and mentioned their Claude enterprise push. Didn't get her email but she's on LinkedIn."
    }
  ],
  response_format: EXTRACTION_SCHEMA
};

// Expected output (guaranteed to match schema)
const expectedOutput = {
  name: "Sarah Chen",
  role: "Head of Product",
  company: "Anthropic",
  email: null,
  confidence: "medium"
};

console.log("STRUCTURED OUTPUTS — Production Extraction Pattern");
console.log("=".repeat(55));
console.log("\nAPI call uses response_format with json_schema:");
console.log("  strict: true — guarantees schema compliance");
console.log("  All fields defined with types and descriptions");
console.log("  additionalProperties: false — no unexpected fields");
console.log("\nExpected output:");
console.log(JSON.stringify(expectedOutput, null, 2));

// API comparison framework
console.log("\n" + "=".repeat(55));
console.log("OPENAI API DECISION FRAMEWORK (2026)");
console.log("=".repeat(55));

const decisions = [
  { api: "Chat Completions", when: "Full control, stateless, any LLM", tradeoff: "You manage state" },
  { api: "Responses API",    when: "Web search, file search, tools, no thread mgmt", tradeoff: "Newer, less documented" },
  { api: "Assistants v2",    when: "Persistent threads, code interpreter, managed files", tradeoff: "Opaque, harder to debug" },
];

decisions.forEach(d => {
  console.log("\n" + d.api);
  console.log("  Use when: " + d.when);
  console.log("  Tradeoff: " + d.tradeoff);
});

console.log("\nStructured Outputs: USE BY DEFAULT for extraction.");
console.log("JSON mode: legacy fallback only.");

Interview question

What’s the difference between Chat Completions, Responses API, and Assistants API? When would you use each?

PM angle

Resources

• DOCS OpenAI Chat Completions API — The industry standard API format. Know it cold.
• DOCS OpenAI Responses API — New in 2025. Web search, file search, tools without thread management.
• DOCS OpenAI Assistants API v2 — Persistent threads and code interpreter. Best for internal tools.
• DOCS Structured Outputs — JSON Schema-constrained output. The default for production extraction — not JSON mode.
• DOCS OpenAI Fine-tuning — Fine-tuning + distillation workflow for GPT-4o → GPT-4o-mini.
• DOCS OpenAI Batch API — 50% cost reduction for async workloads. Verify current limits at docs.