Effective Context for AI: Prompt Engineering

🧠 Context is King

The single biggest factor in the quality of AI output is the quality of the input (context). An LLM is a completion engine. If you give it vague context, it hallucinates. If you give it specific context, it reasons.

The 4 C’s of Context

1. Capacity: The role (Act as a Senior Android Engineer).
2. Context: The background (Project is MVVM, Hilt, Room).
3. Constraints: The rules (Use Kotlin, no Java, handle errors with Result).
4. Chain of Thought: The process (Think step-by-step).

📝 Example: The Bad Prompt

“Create a login screen.”

Result: A generic XML layout, probably using RelativeLayout or LinearLayout, maybe in Java.

🚀 Example: The Good Prompt

“Act as a Senior Android Developer. Create a Login Screen using Jetpack Compose (Material 3).

Context:

• Use Hilt for DI.
• ViewModel should expose StateFlow<LoginUiState>.
• Handle loading, success, and error states.

Constraints:

• Use OutlinedTextField for inputs.
• Validate email format.
• Do NOT use LiveData.

Steps:

1. Define LoginUiState.
2. Create LoginViewModel.
3. Implement LoginScreen composable.”

Result: Production-ready code that fits your architecture.

🛠️ Optimizing Context Windows

Don’t paste 50 files. Be selective.

• Relevant Files Only: Paste the ViewModel and the Repository interface, not the whole data layer.
• Summarize: Instead of pasting a 2000-line file, say “User model has id, name, email.”

🤖 Advanced Technique: Few-Shot Prompting

Show, don’t just tell. Give examples of desired output.

“Convert this JSON to a Kotlin Data Class.

Example: Input: {"id": 1, "name": "John"} Output:

@Serializable
data class User(val id: Int, val name: String)

Now convert this: …”

🏁 Conclusion

Prompt Engineering is the new coding. Writing effective prompts is a skill that separates average developers from 10x AI-augmented engineers.