The Art of Prompt Engineering: Getting the Best from LLMs

Why Prompt Engineering Matters

When I first started using the Gemini API, I would throw vague questions at it and wonder why the responses were mediocre. The breakthrough came when I realised that AI models are only as good as the instructions you give them. Prompt engineering is the skill of crafting those instructions to get consistently excellent results.

It's not just about being more specific it's about understanding how these models think, what context they need, and how to structure your request for maximum clarity.

The Five Prompting Techniques I Use Daily

1. Zero-Shot Prompting

The simplest approach ask the model directly without any examples. This works for straightforward tasks:

prompt = "Explain the difference between supervised and unsupervised learning in 3 bullet points."

Zero-shot works best when the task is well-defined and the expected format is obvious.

2. Few-Shot Prompting

Give the model a few examples of the input-output pattern you want before asking your actual question:

prompt = """
Convert these product descriptions to short taglines:

Description: A waterproof smartwatch with 7-day battery life and heart rate monitoring.
Tagline: "Stay connected. Stay active. All week long."

Description: A noise-cancelling headphone with 30-hour battery and spatial audio.
Tagline: "Silence the world. Hear your music."

Description: A portable solar charger that can power two devices simultaneously.
Tagline:
"""

Few-shot is incredibly powerful because it teaches the model your desired style and format through demonstration.

3. Chain-of-Thought (CoT)

Ask the model to think step by step. This dramatically improves accuracy for reasoning tasks:

prompt = """
A store has 150 items. 40% are electronics, and 25% of electronics are on sale.
How many electronics are on sale?

Think step by step before giving the final answer.
"""

The "think step by step" instruction forces the model to show its work, which catches errors before they compound.

4. Role-Based Prompting

Assign the model a specific role or persona to shape its responses:

prompt = """
You are an experienced Python code reviewer at a top tech company.
Review this code for bugs, performance issues, and best practices.
Be direct and specific. Provide corrected code where needed.

Code to review:
{user_code}
"""

Role assignment gives the model a framework for how to approach the task and what level of expertise to assume.

5. Output Format Specification

Always tell the model exactly how you want the output structured:

prompt = """
Analyse this customer feedback and return your analysis as JSON with these keys:
- sentiment: "positive", "negative", or "neutral"
- key_topics: array of main topics mentioned
- urgency: "high", "medium", or "low"
- suggested_action: one sentence recommendation

Feedback: "{customer_feedback}"
"""

Common Mistakes to Avoid

• Being too vague "Write something about AI" will give generic output. "Write a 200-word blog intro about how Gemini API is being used in Indian startups" gives laser-focused results
• Not specifying length Always include word count, number of bullet points, or paragraph count
• Forgetting context The model doesn't know your project, audience, or constraints unless you tell it
• Not iterating Your first prompt is rarely your best. Refine based on the output you get
• Ignoring temperature settings Lower temperature (0.10.3) for factual/analytical tasks, higher (0.70.9) for creative tasks

My Prompt Engineering Workflow

For any serious project, I follow this workflow:

• Define the goal What exactly should the output look like?
• Draft the prompt Include role, context, task, format, and constraints
• Test with edge cases Try unusual inputs to see if the prompt is robust
• Refine Adjust based on failure patterns
• Template it Once it works, save it as a reusable template with variable slots

The Bottom Line

Prompt engineering is genuinely a skill and one that's becoming more valuable by the day. As AI models get more capable, the gap between a mediocre prompt and a great one grows wider. The people who can reliably extract excellent outputs from LLMs will have a significant advantage in any field.

Start practising with simple tasks, keep a library of prompts that work, and always ask yourself: "What context does the model need to give me exactly what I want?"