DEV Community: Aren Solomon

Fine-tuning and how it works(The Core "Why" and "How")

Aren Solomon — Sun, 12 Apr 2026 07:22:26 +0000

Fine-tuning refers to the process of taking a pre-trained language model and further training it for a specific task or domain-specific dataset. It basically improves the performance of a foundation model by allowing the model to adapt its knowledge and capabilities to better suit the requirements of the business use case.

Use Cases

Let's consider these use cases. If you are working on a task that requires industry knowledge, you can take a pre-trained model and fine-tune it with industry data. If the task involves medical research, for example, the pre-trained model can be fine-tuned with articles from medical journals to achieve more contextualized results.

There are two ways to fine-tune a model;

👉 Instruction fine-tuning, uses a dataset of prompt-response pairs on how the model should respond to a specific instruction.

👉 Reinforcement learning from human feedback (RLHF), provides human feedback data, resulting in a model that is better aligned with human performance.

How fine-tuning works

Step 1: Start with a pre-trained language model that already understands language at a high level.

Step 2: Gather and clean the data relevant to your specific domain.

Step 3:Add task-specif layers

Step 4: Fine-tune the model

Step 5: Evaluate and iterate: Test the model against a "held-out" set of data to ensure it has improved without losing its general capabilities.

Terraform as an IAC Tool for Provisioning Resources

Aren Solomon — Thu, 26 Feb 2026 08:13:16 +0000

The growing demand in business today has led to a cultural shift known as DevOps. This approach highlights the importance of collaboration and communication between software developers and IT operations teams, while also automating software delivery and infrastructure changes.

The days of manually configuring hundreds of servers, networks, and firewall rules through a user interface are behind us. With Infrastructure as Code (IAC), you need to specify the desired end state of your infrastructure. IAC tools will then automatically provision the infrastructure for you.

Terraform is an Infrastructure as Code (IaC) tool developed by HashiCorp that enables users to provision Google Cloud resources using declarative configuration files. With Terraform, you can manage various Google Cloud resources, including virtual machines, containers, storage, and networking. It allows you to express your infrastructure as code in a straightforward and human-readable language known as HashiCorp Configuration Language (HCL).

The Configuration Workflow includes:
👉 Author - Author the configuration files based on the scope.
👉 Initialize - Download the provider plugins and initialize the directory.
👉 Plan - View execution plan for resources created, modified, or destroyed.
👉 Apply - Create actual infrastructure resources.

Use Terraform for Google Cloud to:
✅ Provision resources, which means you can use resource blocks to define infrastructure elements such as VMs, network, and firewall resources.

✅ Create resource dependencies: You can create explicit dependencies between resources so that a given resource can only be created after the creation of another resource.

✅ Standardize configurations: You can standardize how a given resource is created by creating modules.

✅ Validate inputs to resource arguments: You can limit the values that a user can provide for a given resource argument using validation rules within Terraform.

DevOps #Terraform #Cloud

Choose Serverless, Choose Cloud Run Today As A Developer or A Cloud Customer!

Aren Solomon — Mon, 23 Feb 2026 17:39:12 +0000

Cloud Run is a developer-focused, fully managed application platform for running code on Google's scalable infrastructure.

Designed to make developers productive and allow them to focus on solving business problems. As a serverless offering, it abstracts away all infrastructure management, allowing developers to focus on Code.

Cloud Run increases innovation velocity and reduces cost and risk. It makes it an ideal solution for quickly and efficiently deploying web applications, APIs, and event-driven microservices.

It leverages Cloud Build to build container images, and works well together with other services on Google Cloud, so you can build full-featured applications.

Cloud Run has use cases across:

👉 Public and Private Services: Websites, REST, or GraphQL API.

Streaming with WebSockets, AI agents, private HTTP, or gRPC microservices.

👉 Data Processing: Process queue messages, event-driven architecture, scheduled scripts, Background processing, Batch data processing.

You can deploy code to Cloud Run Via:

✅ Container: Deploy a container image.

Supported Repositories: Artifact Registry, Docker Hub

✅ Repository: Continuously deploy code from a connected source git repository.

Supported Repository: Github

✅ Source Code: Deploy your code to Cloud Run with a single command.

Supported languages: Python, Node.js, Go, Java, Ruby, PHP, and .NET

✅ Function: Deploy single-purpose functions.

Supported languages: Python, Node.js, Go, Java, Ruby, PHP, and .NET

There are two main resources for executing workloads in Cloud Run, which are Cloud Run Services and Cloud Run Jobs:

Cloud Run Services is used to execute code that responds to web requests or events.

The services include:

✅ Serving HTTP traffic under a dedicated endpoint

✅ Pay when code is processing requests.

✅ Use out-of-the-box URL with TLS

✅ Leverage automatic Scaling with minimum and maximum instances.

✅ Access a dedicated endpoint that is invoked by HTTPS request events, websockets, HTTP/2, and gRPC.

✅ Empty built-in traffic splitting for gradual rollouts and rollbacks.

✅ Utilize revision history.

Cloud Run Job is used to execute code that performs work on jobs and exits when finished.

Jobs execute tasks to completion, and these features:

✅ Do not have an HTTP endpoint

✅ Have a maximum task timeout

✅ Are executed manually or on a schedule.

✅ Run a specified number of tasks up to seven days

✅ Are paid only while the job is executing

✅ Have execution history.

A developer team might choose Cloud Run for their application because:

👍 They want to deploy containerized applications without the burden of managing infrastructure or services.

👍 Their application experiences variable traffic, and they want automatic scaling (including scaling down to zero) for cost-efficiency.

👍 They need to run stateless web services, API, or event-driven functions packaged as containers.

DevOps #Cloud

Google Generative AI's Capabilities

Aren Solomon — Thu, 19 Feb 2026 18:13:17 +0000

Have you ever wondered Google's Generative AI's Capabilities?

With Google Cloud tools like:

✅️ Vertex AI
✅️ Search and Conversation
✅️ BigQuery
✅️ Looker

GenAI can be embedded into websites and transform user experience in three ways:

✅️ Streamline user content generation
✅️ Enhance web navigation
✅️ Improve self-service

Streamline content generation

Use Vertex AI to:

👉 Create and translate the content using the translation API.

👉 Generate images from text prompts using Imagen on Vertex AI.

👉 Generate captions for product description from images.
Generate natural languages text content (E.g blog posts) using the Gemini API

Use Gemini for Google workspace to write and refine content for web copy.

Improve content with customer 360 for BigQuery.

Enhance web Navigation

👉 Use Vertex AI Agent Builder to:
Create search and recommendation apps for your website.

👉 Generate grounded responses to complex queries on your enterprise data.

Improve Self-Service

👉 Use Vertex AI search to provide results grounded from enterprise data.

👉 Use Dialogflow with generators to create website chatbots with personalized context aware content.

DevOps #AIEngineer ##GoogleCloud

Retrieval-Augmented Generation

Aren Solomon — Mon, 16 Feb 2026 10:50:10 +0000

There is a significant data gap in standard Large Language Models (LLMs), because these models are frozen at their "knowledge cutoff" date, and the major challenges are in:

✅️ Stale Information ✅️ Lack of Provenance & ✅️ Trainings missing out on private or real-time information.

A Retrieval-Augmented Generation (RAG) system addresses these gaps by turning the LLM into an "open-book" researcher. Instead of relying solely on memory, the system looks up external facts before generating a response.

In a modern RAG architecture, the "Knowledge Base" isn't just one giant pile of text; it is strategically categorized by how often the information changes.

It has two pillars:

✅️ Static Content (The Foundation),
which are Information that remains constant over long periods (e.g., historical archives, core company policies, or product manuals).
Serves as the "Library."

✅️ Dynamic Content (The Real-Time Layer),
Data that updates frequently—sometimes by the second (e.g., live stock prices, current inventory, or the latest news). This solves "knowledge lag."

Top-tier RAG systems use Agentic Workflows or APIs to pull this "Fresh Stuff" on demand. Rather than waiting for a database to re-index, the AI makes a live call to ensure the answer is accurate right now.

Why this matters in 2026
Modern RAG has evolved beyond simple text searching. We now use

GraphRAG in Connecting dots between different pieces of information using "knowledge graphs" to understand complex relationships.

Source Attribution with every claim the AI makes, backed by a direct link to the source, makes the system auditable and trustworthy.

Again, Hybrid Retrieval combines the reliability of the "Static Library" with the speed of "Dynamic APIs" to create a truly intelligent assistant.

Build For The Cloud

Aren Solomon — Fri, 23 Jan 2026 16:01:32 +0000

Hello! I am excited to join the community and eager to collaborate and learn on this journey.