Prompt Engineering

All You Need To Know About Prompt Engineering¶

1. Be precise and explicit¶

Don’t assume hidden meaning. You must spell out every detail: format, tone, constraints.

2. Structure your prompts carefully¶

Start with clear roles: "You are a Technical Writing Assistant…"
Define objective, format, and negative constraints explicitly: *“List 3 best-performing Q1 2025 products, then give 5 strategic bullets, don’t use paragraphs.”.

3. Use agentic workflows¶

Turn your LLM into an autonomous agent:

Remind it it's a multi‑message agent (“don’t stop until fully solved”).
Encourage planning + reflection between tool/API calls.
Insist on tool‑calls, not guesses

4. Utilize context adeptly¶

Feed long inputs and place essential prompts at beginning and end.
Remind it of its role mid‑doc to maintain context.

5. Guide reasoning with chain-of-thought¶

Despite not being a pure reasoning model, LLMs performs better when asked to "think step‑by‑step." Phrasing like “First X, then Y” yields better accuracy on logical tasks.

6. Few-shot / examples¶

Show desired output patterns. Including examples, that helps ensure structure, tone, and format are followed.

7. Iterate empirically¶

Treat prompt tuning like debugging:

Try basic version
Evaluate
Adjust one element
Repeat Empirical tweaks are key.

Sample Prompt Template¶

Markdown
You are a data extractor.
Extract the following fields:
- Title
- Author
- Date
- Summary

Output in JSON.
[TEXT]

Markdown
1 2 3	`You are a math tutor. Solve the problem step by step and explain your reasoning. Problem: [PROBLEM]`

Types of Prompting¶

✦ Zero-shot¶

✦ Few-shot¶

✦ Chain-of-Thought (CoT)¶

✦ ReAct Prompting¶

✦ Tree of Thoughts (ToT)¶

LLM Tuning¶

✅ Use temperature:

0 for deterministic answers¶

>0.7 for creativity¶

ReAct Prompting Example¶

Markdown

You are an intelligent agent that reasons step by step and uses tools.

Question: What is 23 * 47?

Thought: I should calculate step by step.
Action: Multiply(23, 47)

Observation: 1081

Answer: 1081

Tree of Thoughts (ToT) Example¶

Markdown
You are solving: How to reduce energy consumption in a data center by 30% in 6 months.

Generate 3 different high-level strategies.
For each, list pros and cons.
Then pick the most promising one to elaborate.

Strategies:
1. Migrate to cloud.
2. Optimize cooling.
3. Use renewable energy.

...

[Model continues]

✅ Role Prompting:

"You are a senior full-stack developer and technical architect..."

✅ Zero-shot instructions:

Clear goals for each part of the system.

✅ Few-shot examples:

Provide a sample output example.

✅ Chain-of-Thought:

"Explain your reasoning step by step."

✅ ReAct:

"Thought → Action → Observation."

✅ Tree of Thoughts:

"Generate multiple alternatives and evaluate them."

✅ Constraints:

Next.js 15, Tailwind.

✅ Format Control:

Markdown and code blocks.

✅ Proactive Edge-Case Handling:

"Highlight edge cases, scalability concerns, and performance optimizations."

Summery¶

🔹 Clarity: Be precise and explicit. Vague prompts = vague answers.

🔹 Context: Give the model everything it needs to know: background, tone, examples.

🔹 Constraints: Specify format, length, style, and detail level.

🔹 Examples: Provide examples if you want structure or style imitation.

🔹 Iteration: Tweak, test, and compare variations.