We wrapped up Week 1 by diving into Prompt Engineering. A lot of people think prompt engineering is just typing random sentences into a chat box until something works. But when you are building actual software, you have to be structured and deliberate.
1. The RTS Framework
We practiced using the RTS formula: Role, Task, and Style. Instead of saying 'explain Python,' you tell the model: 'You are an expert Python mentor (Role). Explain how dictionaries work with code examples (Task). Keep it concise, friendly, and formatted in bullet points (Style).' The difference in output quality is night and day.
2. Temperature and Token Mechanics
Next, we explored hyperparameters like Temperature. Think of temperature as a creativity knob. If you set it low (like 0.2), the model picks the most probable, factual words—which is what you want for coding, math, or legal answers. If you dial it up to 0.8 or 1.0, it gets more imaginative and varied, which is great for creative writing. We also looked at tokens—chunks of words that models process—and why context limits matter.
3. What is RAG Anyway?
Rashmi introduced RAG (Retrieval-Augmented Generation). It sounds intimidating, but the concept is brilliant: instead of trying to retrain an entire massive model on your private notes, you search your documents for the relevant paragraphs, paste them into the prompt behind the scenes, and tell the AI: 'Answer the user's question using ONLY these notes.' It stops the AI from making things up (hallucinating) and gives you accurate answers every time.