Prompt Engineering: What I Learned About Getting Better Answers from Gen AI

Bella — Wed, 21 Oct 2026 09:27:33 +0000

Generative AI: what is it?

The biggest challenge in using such a powerful tool as generative AI is learning how to get better answers, and how that process can be automated in some way.

I've been studying some ways to do that, and I came across prompt engineering. I really like the approach I found in the Google AI Prompt Essentials course.

Generative AI (or GenAI, for short) is software that takes a prompt, understands your request, and generates a response. It can create text, summaries and emails, answers to questions, reports and insights, computer code, images, translations, and call summaries.

Generative AI sits under the AI umbrella:

Artificial Intelligence → Generative AI → Large Language Models (LLMs) → tools such as ChatGPT

The consequence of a bad prompt

This is a gray zone. Every time you create a different prompt, you can receive a very different output. If you simply ask, "Generate a script to tell my team how to sell more," that's very vague. Which team? Sell what? What information should it use? Do you have references?

The GenAI will return a script, but it will be very poor and unfocused, because it doesn't know what you really need. That's why prompt engineering is so important.

The framework

While studying prompt engineering on Google Digital Skills, I learned a good framework that helped me get better outputs. Basically, you write your prompt based on:

Task + Context + References + Evaluate + Iterate

Evaluate: It's critical to actually read the output and evaluate whether it's coherent and heading in the direction you want for your idea.

GenAI can have hallucinations: when it returns a very confident answer that is incorrect or made up. This can happen because AI works by prediction. It's likely to produce whatever comes next based on patterns, but it doesn't "know" everything. Be careful!

Main takeaways

• A framework helps, but you can also explore other ways to improve your prompt.
• Don't rely on outputs with your eyes closed.
• Refined prompt, better results!

Kubernetes Overview

Bella — Fri, 19 Jan 2024 14:50:03 +0000

What is Kubernetes?

Kubernetes (K8s) is an open-source container orchestration tool developed by Google. It facilitates the management of containerized applications across various deployment environments, including physical machines, cloud, and virtual machines.

Problem Kubernetes Solves:

The shift from monolith to Microservices involves encapsulating applications/services in multiple or thousands of containers, necessitating effective container management. Monoliths represent a single, large application, while Microservices break it down into smaller, independent services.

Features of Orchestration Tools:

High availability or no downtime
Scalability for optimal performance
Disaster recovery with easy backup and restore

Kubernetes Architecture:

A Kubernetes cluster comprises at least one master node (virtual or physical) and several worker nodes. Key components in the master node include the API Server, Controller Manager, Scheduler, and Etcd for maintaining the cluster's current state.

Worker nodes run applications, and the virtual network enables communication between nodes.

Main Kubernetes Components:

Node and Pod:

Node: Virtual or physical machine
Pod: An abstraction of a container, with each pod typically containing a single application.
Pods communicate using IP addresses and are ephemeral.

Service:

A component with a permanent IP, ensuring consistency even if pods change. External services communicate externally, while internal services are the default type.

Ingress:

Positioned before services, it creates a dedicated load balancer for Kubernetes, enhancing security. It allows the use of domains instead of IP-like numbers.

ConfigMap and Secret:

ConfigMap: ConfigMaps in Kubernetes provide a means to inject configuration data into pods of an application.
Secret: Similar to ConfigMap but used for storing sensitive data, encrypted for security.

Volume:

Physical storage on local or remote machines, ensuring data persistence even if applications restart.

Deployment:

A blueprint for 'my-app' pods, defining the number of replicas. Ideal for stateless apps, providing easy scalability.

Example of deployment

apiVersion: apps/v1
kind: Deployment
metadata:
  name: nginx-deployment
  labels:
    app: nginx
spec:
  replicas: 3
  selector:
    matchLabels:
      app: nginx
  template:
    metadata:
      labels:
        app: nginx
    spec:
      containers:
      - name: nginx
        image: nginx:1.14.2
        ports:
        - containerPort: 80

Distributed Systems and Containers:

Replicating applications becomes simpler with permanent IPs, avoiding constant adjustments when pods change. Load balancing directs requests to less busy pods.

Database Considerations:

Databases, due to state/data, can't be replicated via deployment. StatefulSet, designed for stateful apps or databases, manages pods that can update data. Hosting databases outside the cluster is a recommended approach for consistency.

*References: *

TechWorld with Nana

kubernetes documentation

DEV Community: Bella