Home › Day 9
Day 9 · 1 Hour

LLM & Prompt Engineering

Groq API, system prompts, temperature — talk to an LLM the way engineers do, not users.

0–5 min
Real Hook
5–25 min
Core Concept
25–45 min
Live Coding
45–55 min
You Code It
55–60 min
Cheat Sheet
🔥 Real-World Hook (0–5 min)

Two people ask ChatGPT to write a Python function. One gets a 10-line vague answer. The other gets a production-ready function with docstring, type hints, and edge case handling. Same model. Different prompt.

Prompt engineering is the new skill gap. Companies pay ₹15–25 LPA for engineers who know how to instruct LLMs reliably. Today you learn the difference.

📈 Prompt Engineering Principles (5–25 min)

TechniqueWhat it doesExample
System promptSets personality & constraints"You are a senior Python developer..."
Few-shotShow examples before the taskInput/output pairs
Chain-of-thoughtAsk for reasoning steps"Think step by step"
Temperature 0Deterministic outputCode, JSON, facts
Temperature 0.7+Creative outputStories, brainstorming

💻 Live Demo — Code Review Bot (25–45 min)

>from groq import Groq
import os

client = Groq(api_key=os.getenv("GROQ_API_KEY"))

# System prompt defines the LLM's role and behaviour
SYSTEM = "You are a senior Python code reviewer at a top tech company.
Review code for: correctness, PEP8 style, edge cases, performance.
Be specific. Rate each issue: Critical / Warning / Suggestion.
Keep response under 200 words."

code_to_review = "
def get_user(users, id):
    for u in users:
        if u["id"] == id:
            return u
    return None
"

response = client.chat.completions.create(
    model="llama3-8b-8192",
    temperature=0,                          # deterministic for code review
    messages=[
        {"role": "system", "content": SYSTEM},
        {"role": "user",   "content": f"Review this Python code:\n{code_to_review}" }
    ]
)
print(response.choices[0].message.content)

# Few-shot: teach the model the exact output format
few_shot_msgs = [
    {"role": "user",      "content": "Classify: Free iPhone click here!!!"},
    {"role": "assistant", "content": "SPAM"},
    {"role": "user",      "content": "Classify: Meeting at 3pm tomorrow"},
    {"role": "assistant", "content": "HAM"},
    {"role": "user",      "content": "Classify: Congratulations you won a prize!"},
]

📋 You Code It (45–55 min)

✍ Task

Write a prompt that converts a plain English description into a Python function signature with type hints. Test it on: "a function that takes a list of names and returns them sorted with duplicates removed". Try temperature 0 vs 0.9 — observe the difference.

⚡ Day 9 Cheat Sheet

Groq(api_key=os.getenv(...))
Create Groq client
client.chat.completions.create()
Call the LLM
messages=[{role, content}]
Conversation turns
role: "system"
Set personality/constraints
temperature=0
Deterministic output
temperature=0.7
Creative output
response.choices[0].message.content
Get text response
max_tokens=200
Limit response length
← Day 08Day 10 · LangChain →