Prompt Engineering

Better way of Framing an input for GenAI model to get desired output by reducing the cost

input/prompt ——> AI model ——> desired output

Good prompt significantly reduce the cost?

• Get Kubernetes deployment/service → Python → Generate

• Prepare prompt

• API call to LLM (OpenAI, Deepseek)

  • (API calls are not free)

• Bad prompt/input generates more tokens/words

• Number of tokens(response words) increases price

Zero Shot prompting:

Provide a prompt without any example

• general version:Generate a kubernetes manifest for deployment resource

• better version:Generate ONLY kubernetes manifest for deployment resource

Few Shot prompting:(recommended)

provide examples

• Example: Create a name which starts with N

• Answer: Name for alphabet N is Nani

• Example: Fetch the Docker version

• Answer:

    #!/bin/bash
  
    #######
    # Author:
    # version
    # date
    #########
  
    docker --version
  

• Fetch the versions of all the processes that are installed on the machine

Multi shot prompting:

provide more examples

COT (chain of thoughts):

• Enhances the performance of LLM (Reasoning)

• using chain of thoughts we can derive best results from LLMs with resoning capability

Strategy to use prompt. engineering:

• Provide context

• Provide the instruction

• Provide examples (few shot prompting)

• Define output format