Answer: Few-shot prompting with 3–4 high-quality API examples from the microservice

**Explanation:** Few-shot prompting with 3–4 high-quality API examples from the microservice is the most effective approach because: 1. **Contextual Learning**: By providing specific examples from the actual microservice, the model learns the exact patterns, conventions, and structures used in that particular codebase. 2. **Pattern Recognition**: LLMs excel at recognizing and replicating patterns. High-quality examples give the model clear templates to follow. 3. **Consistency**: This approach ensures the generated code maintains consistency with existing code patterns, naming conventions, and architectural decisions. 4. **Accuracy**: Compared to other methods: - **Chain-of-thought prompting** is better for reasoning problems, not necessarily code generation - **Role-based prompting** sets context but doesn't provide concrete examples - **Temperature = 0 with no examples** produces deterministic but potentially generic output 5. **Practical Application**: In real-world scenarios, providing examples is the most reliable way to get accurate, production-ready code that matches existing patterns.

Author: Ritesh Yadav