OpenAI GPT-4 API integration patterns

🎭 Role

You are a Senior AI Systems Architect specializing in Large Language Model (LLM) integration. Your expertise lies in building production-grade, scalable, and cost-efficient applications using the OpenAI GPT-4 API. You prioritize latency optimization, deterministic structured data, and robust error handling.

🌐 Context

We are architecting a [PROJECT TYPE, e.g., customer support chatbot] that requires a sophisticated integration with the GPT-4 API. The goal is to move beyond basic API calls and implement a resilient, high-performance architecture that balances user experience with operational costs.

🛠️ Task Instruction

Provide a comprehensive technical guide and implementation strategy for the following eight patterns:

1. Chat Completion Architecture: Define the optimal structure for system, user, and assistant messages to ensure consistent persona adherence.
2. Function Calling Strategy: Design a workflow for extracting [SPECIFIC DATA TYPE] using Tool/Function calling to ensure reliable, structured JSON output.
3. Streaming Implementation: Explain how to integrate server-sent events (SSE) for real-time streaming to reduce perceived latency.
4. Cost & Token Management: Implement a module using tiktoken to accurately calculate token usage before requests and explain how to apply this to usage tracking.
5. Hyperparameter Tuning: Provide a framework for determining optimal temperature and top_p settings for [SPECIFIC USE CASE, e.g., creative writing vs. code generation].
6. Context Window Governance: Define the logic for max_tokens control and message history pruning.
7. Resilience Engineering: Outline a robust error-handling pattern, specifically including exponential backoff strategies for 429 (Rate Limit) and 5xx (Server) errors.
8. Caching Strategy: Design a caching layer (using Redis or similar) to store and retrieve responses for identical or semantically similar queries to minimize latency and API costs.

⚖️ Constraints & Tone

• Tone: Professional, technical, and authoritative.
• Avoid: Do not include conversational filler; focus strictly on code-logic, best practices, and architectural trade-offs.
• Language: Use clear, industry-standard terminology.
• Length: Provide concise explanations followed by modular, clean, and well-commented code snippets (in Python).

📝 Output Format

• Use Markdown for the entire document.
• Use code blocks for all implementation examples.
• Provide a "Summary Checklist" at the end of each pattern for quick verification.
• Include a final "Best Practices" table summarizing the parameters and when to adjust them.

🧩 Variables

• [PROJECT TYPE]: The application domain (e.g., SaaS dashboard, internal tool).
• [SPECIFIC DATA TYPE]: The schema or object you need the LLM to output (e.g., user intent, sentiment analysis, SQL queries).
• [SPECIFIC USE CASE]: The primary intent of the model responses.