<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Clifford E. D&#39;Souza on Medium]]></title>
        <description><![CDATA[Stories by Clifford E. D&#39;Souza on Medium]]></description>
        <link>https://medium.com/@cliffordedsouza?source=rss-b2feb9ac956------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*V0TUCCGCaQvbjtgdDKwJJg.jpeg</url>
            <title>Stories by Clifford E. D&amp;#39;Souza on Medium</title>
            <link>https://medium.com/@cliffordedsouza?source=rss-b2feb9ac956------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 16:30:09 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@cliffordedsouza/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Create a New Claude Code Plugin]]></title>
            <link>https://cliffordedsouza.medium.com/create-a-new-claude-code-plugin-a4f0425cd586?source=rss-b2feb9ac956------2</link>
            <guid isPermaLink="false">https://medium.com/p/a4f0425cd586</guid>
            <category><![CDATA[claude-code]]></category>
            <category><![CDATA[anthropic-claude]]></category>
            <category><![CDATA[claude-code-plugins]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[coding]]></category>
            <dc:creator><![CDATA[Clifford E. D'Souza]]></dc:creator>
            <pubDate>Sun, 06 Sep 2026 15:38:53 GMT</pubDate>
            <atom:updated>2026-09-06T15:38:53.043Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OzFrPLgeoBuD8gCvFCMYFw.png" /></figure><p>A <strong>Claude Code plugin</strong> is a packaging format used to bundle and distribute custom capabilities for Claude. It is not a distinct behavior or feature itself, but rather a structured directory that can contain a combination of skills, agents, slash commands, and tool connections (MCP servers) so they can be installed and shared as a single unit.</p><h3>Plugin, Skill and Agent</h3><p>Before going further, let us first understand the differences between a plugin, agent and skill. Table below provides a quick comparison.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/769/1*BTBvPoVz-mKdaNndSGraMw.jpeg" /></figure><h3>Official Skill Creator Plugin</h3><p>We’ll use the official <a href="https://github.com/anthropics/claude-plugins-official/tree/main/plugins/skill-creator">skill-creator plugin</a> from Antrophic Marketplace to create the new plugin. From <a href="https://code.claude.com/docs/en/quickstart/">Claude code</a> CLI, we will run the below instruction to install the plugin, since it is not available by default.</p><pre>/plugin install skill-creator</pre><p>After restarting <a href="https://code.claude.com/docs/en/quickstart/">Claude code</a> the plugin can be triggered. Here’s what I see when triggering the <a href="https://github.com/anthropics/claude-plugins-official/tree/main/plugins/skill-creator">skill-creator plugin</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tpm7vypt9wjbJsE8ZCuAkA.jpeg" /></figure><p>The <a href="https://github.com/anthropics/claude-plugins-official/tree/main/plugins/skill-creator">skill-creator plugin</a> is a meta-tool designed to help you scaffold, build, and test <a href="https://code.claude.com/docs/en/quickstart/">Claude code</a> capabilities—including full plugins—through an interactive, data-driven workflow. Instead of manually writing markdown files and manifests from scratch, it acts as a wizard that guides you through the architecture and generates the necessary files.</p><p>Here are the broad steps to create the new plugin. I will select option “4. I’ll describe it”:</p><p>Tell Claude what you want to build to trigger the plugin’s creation workflow.<br>Prompt example: “<strong><em>I want to build a new plugin to format my source code per the Google Java Style Guide. Please use skill-creator to help me design and generate it.</em></strong>”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*s7QV7382q4wuwp3dUq4PhQ.jpeg" /><figcaption>Prompting Claude to build the new plugin</figcaption></figure><p>Answer the discovery questions asked by <a href="https://code.claude.com/docs/en/quickstart/">Claude code</a> / <a href="https://github.com/anthropics/claude-plugins-official/tree/main/plugins/skill-creator">skill-creator plugin</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UsVlOgtfQw5_Wmvl0BA39g.jpeg" /><figcaption>Answering the discovery questions from Claude code</figcaption></figure><p>The skill-creator will start an interactive discovery process to plan your plugin’s architecture. It will ask you:</p><ul><li><strong>The core problem:</strong> What specific workflow or task the plugin solves.</li><li><strong>The components needed:</strong> Whether your plugin needs slash commands (for manual invocation), background skills (for automatic context), subagents (for isolated, complex reasoning), or MCP server integrations.</li><li><strong>Edge cases &amp; Logic:</strong> How it should handle empty inputs, missing files, or specific formatting rules.</li></ul><p>After answering all the discovery related questions, <a href="https://code.claude.com/docs/en/quickstart/">Claude code</a> is ready and requires confirmation from the user to proceed. Here’s a screen of what that looks like:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IeuTeCCfmYT4CeDKbDEkZA.jpeg" /></figure><p>I noticed that <a href="https://code.claude.com/docs/en/quickstart/">Claude code</a> self-checks and tests the output to make sure the plugin is working correctly. Here’s the action in the <a href="https://code.claude.com/docs/en/quickstart/">Claude code</a> CLI:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*k8OSSHfOMWJ8sOaLKHSIXg.jpeg" /><figcaption>Claude code testing its ouput</figcaption></figure><p>After answering the discovery questions and the requirements are clear, the tool will automatically generate the standard plugin directory structure and templates. This typically includes:</p><ul><li>plugin.json: The manifest file defining your plugin&#39;s metadata and dependencies.</li><li>commands/: Markdown files for any custom slash commands.</li><li>skills/: The SKILL.md files containing the system instructions and YAML frontmatter.</li><li>agents/: Persona and context definitions if you are spawning subagents.</li></ul><h3>Iterations — Core to Plugin Development</h3><p>Before finalizing the new plugin, there’s a human review step included. This gives an opportunity to tweak any decisions that would be finally embedded along with the plugin. From the screen shot below, you can see that <a href="https://code.claude.com/docs/en/quickstart/">Claude code</a> CLI using the skill is asking for manual review and confirmation of the output. It is doing so as there would be some breaking changes. This needs confirmation on whether it is okay to proceed with the code edits. It also does a comparison between skill vs no skill (basic <a href="https://code.claude.com/docs/en/quickstart/">Claude code</a> base knowledge). The skill version gives better results — asking for confirmation before making breaking code changes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*19kXZhYZeLVi6aXsASiw6w.jpeg" /></figure><p>For the purpose of performing the manual reviews, <a href="https://code.claude.com/docs/en/quickstart/">Claude code</a> launches a local browser site or HTML viewer and shows the output. These are the results of the plugin under test working on a randomly generated codebase that is intentionally generated ‘un-tidy’. The output summarizes the list of safe edits performed as well as those requiring further manual decision inputs on how to proceed. The answers provided here would then be used to perform the final code edits.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YcCWc3kp9P5w_Y-jU7zagg.jpeg" /></figure><p>This output also has insightful comments from <a href="https://code.claude.com/docs/en/quickstart/">Claude code</a> on what actions it performed. Here’s an example:</p><pre>I ran google-java-format (v1.x jar, JDK 21) over `src/main/java`, then a style pass for the rules the formatter cannot apply, then compiled the result with `javac` to confirm nothing broke. `build/generated/Gen.java` and `build.gradle` were left alone.</pre><p>After reviewing all the changes and providing the feedback, it is now time to switch back to <a href="https://code.claude.com/docs/en/quickstart/">Claude code</a> and prompt that you’re done with reviewing. <a href="https://code.claude.com/docs/en/quickstart/">Claude code</a> will then implement the code edits and re-run the code scan. This goes on for a few iterations until we don’t see any ‘decision required’ entries from the eval output in the eval viewer pages.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*J4UC9PhSeL1oGnHkszyUpQ.jpeg" /></figure><h3>Preparing the Plugin for the Marketplace</h3><p>In <a href="https://code.claude.com/docs/en/quickstart/">Claude code</a> CLI, I provided this prompt: “<strong><em>package the plugin for the anthropic marketplace</em></strong>”. The plugin is now ready for submission to the Anthropic marketplace.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zYZRDWyRdmZ_C-96ndUmdQ.jpeg" /></figure><h3>Conclusion</h3><p>Building custom plugins for <strong>Claude Code</strong> bridges the gap between manual prompts and automated, scalable workflows. By leveraging the meta-capabilities of the <strong>skill-creator plugin</strong>, you can bypass the tedious process of writing manifests and markdown files from scratch. Instead, an interactive, discovery-driven workflow handles the heavy lifting—from architecture design to generating files like plugin.json.</p><p>The true power of this process lies in its <strong>iterative nature</strong>. As we saw during the manual review phase, Claude Code doesn’t just generate code and hope for the best; it self-checks, runs localized evaluations, and actively asks for user validation before introducing potential breaking changes. This human-in-the-loop approach ensures that by the time you run the final packaging prompt for the <strong>Anthropic Marketplace</strong>, your plugin is highly optimized, stable, and ready for distribution.</p><p><strong>💡 Key Takeaways for the Road</strong></p><ul><li><strong>Plugins are Bundles, Not Behaviors:</strong> Remember that a plugin is a structured packaging format used to distribute a unified mix of skills, agents, slash commands, and MCP servers.</li><li><strong>Let the Wizard Do the Scaffolding:</strong> Always utilize the skill-creator meta-tool to architect your plugins efficiently through automated discovery questions.</li><li><strong>Embrace the Iteration Cycle:</strong> Use the local browser evaluation viewer to review edits and address ‘decision required’ flags. Iteration is a core step to achieving flawless plugin logic before marketplace submission.</li></ul><p><strong>Thank you for reading!</strong> If you found this guide helpful in kickstarting your Claude Code development journey, please <strong>show your appreciation by giving this article a few claps 👏</strong> — it helps more developers find this content!</p><p>💬 <strong>Over to you:</strong> Are you planning to build a custom plugin, or are you currently stuck on a specific discovery setup? <strong>Drop your questions and thoughts in the comments section below</strong>, and let’s discuss!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a4f0425cd586" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Building a Data Analytics Pipeline on AWS]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://cliffordedsouza.medium.com/building-a-data-analytics-pipeline-on-aws-dbf4ac5881ad?source=rss-b2feb9ac956------2"><img src="https://cdn-images-1.medium.com/max/1584/1*c3FJdY_gipCHSnNjujIXGA.png" width="1584"></a></p><p class="medium-feed-snippet">Background</p><p class="medium-feed-link"><a href="https://cliffordedsouza.medium.com/building-a-data-analytics-pipeline-on-aws-dbf4ac5881ad?source=rss-b2feb9ac956------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://cliffordedsouza.medium.com/building-a-data-analytics-pipeline-on-aws-dbf4ac5881ad?source=rss-b2feb9ac956------2</link>
            <guid isPermaLink="false">https://medium.com/p/dbf4ac5881ad</guid>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[datapine]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[aws]]></category>
            <category><![CDATA[data-engineering]]></category>
            <dc:creator><![CDATA[Clifford E. D'Souza]]></dc:creator>
            <pubDate>Sun, 23 Aug 2026 18:58:00 GMT</pubDate>
            <atom:updated>2026-08-24T16:03:05.369Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Setting up a Data Analytics Workflow using Claude Code]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://cliffordedsouza.medium.com/setting-up-a-data-analytics-workflow-using-claude-code-74b38b2f1bfd?source=rss-b2feb9ac956------2"><img src="https://cdn-images-1.medium.com/max/2600/1*zB3RWARbnpj8V0MXNVfbhw.png" width="2816"></a></p><p class="medium-feed-snippet">I wanted to explore Claude Code, a terminal based IDE for Agentic AI interfacing, from a data analytics workflow setup dimension. Since I&#x2026;</p><p class="medium-feed-link"><a href="https://cliffordedsouza.medium.com/setting-up-a-data-analytics-workflow-using-claude-code-74b38b2f1bfd?source=rss-b2feb9ac956------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://cliffordedsouza.medium.com/setting-up-a-data-analytics-workflow-using-claude-code-74b38b2f1bfd?source=rss-b2feb9ac956------2</link>
            <guid isPermaLink="false">https://medium.com/p/74b38b2f1bfd</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[claude-code]]></category>
            <category><![CDATA[aws]]></category>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Clifford E. D'Souza]]></dc:creator>
            <pubDate>Sun, 16 Aug 2026 18:58:12 GMT</pubDate>
            <atom:updated>2026-08-16T18:58:12.924Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[How to Build Autonomous AI Engineers with Claude Code: A Complete Illustrated Guide]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://cliffordedsouza.medium.com/how-to-build-autonomous-ai-engineers-with-claude-code-a-complete-illustrated-guide-e06469903d53?source=rss-b2feb9ac956------2"><img src="https://cdn-images-1.medium.com/max/1600/1*Mue4EG1oD16Hms0xDK5s0A.png" width="1600"></a></p><p class="medium-feed-snippet">Move beyond basic code completion. Learn how to construct self-correcting agentic loops, custom skills, and MCP integrations with&#x2026;</p><p class="medium-feed-link"><a href="https://cliffordedsouza.medium.com/how-to-build-autonomous-ai-engineers-with-claude-code-a-complete-illustrated-guide-e06469903d53?source=rss-b2feb9ac956------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://cliffordedsouza.medium.com/how-to-build-autonomous-ai-engineers-with-claude-code-a-complete-illustrated-guide-e06469903d53?source=rss-b2feb9ac956------2</link>
            <guid isPermaLink="false">https://medium.com/p/e06469903d53</guid>
            <category><![CDATA[anthropic-claude]]></category>
            <category><![CDATA[claude-code]]></category>
            <category><![CDATA[coding]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Clifford E. D'Souza]]></dc:creator>
            <pubDate>Mon, 20 Jul 2026 18:16:48 GMT</pubDate>
            <atom:updated>2026-07-20T18:31:22.184Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[How Claude Code helped me deploy to AWS]]></title>
            <link>https://cliffordedsouza.medium.com/how-claude-code-helped-me-deploy-to-aws-dc4c7ce6b59a?source=rss-b2feb9ac956------2</link>
            <guid isPermaLink="false">https://medium.com/p/dc4c7ce6b59a</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[claude-code]]></category>
            <dc:creator><![CDATA[Clifford E. D'Souza]]></dc:creator>
            <pubDate>Sun, 21 Jun 2026 17:25:06 GMT</pubDate>
            <atom:updated>2026-06-22T03:51:29.747Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*xMJEXxpMpRh6RI4U" /><figcaption>Photo by <a href="https://unsplash.com/@brechtcorbeel?utm_source=medium&amp;utm_medium=referral">Brecht Corbeel</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>Background</h3><p>With <a href="https://claude.com/product/claude-code">Claude Code</a>, anyone can create an application running on the local machine. A prompt such as ‘Write me an app to do ***replace with some functionality***’ will get you started. I did the same thing and created an application that would accept an innovative idea from the user and then provide an evaluation of the idea from a business value perspective. In this article I would like to share my experience in getting this application to run on AWS infrastructure.</p><h3>Implementation</h3><p>Since the application is AI enabled, an Anthropic API key would be required. You can get one from the <a href="https://platform.claude.com/">Claude Console</a>. I also used an AWS free tier account for my deployment needs. Before submitting any prompt to <a href="https://claude.com/product/claude-code">Claude Code</a>, make sure the following CLI tools are installed locally. <a href="https://claude.com/product/claude-code">Claude Code</a> takes advantage of CLI tools heavily:</p><ul><li><a href="https://docs.aws.amazon.com/cli/latest/userguide/getting-started-install.html">AWS CLI tool</a></li><li><a href="https://cli.github.com/">GitHub client tool</a></li><li><a href="https://docs.docker.com/engine/install/">Docker</a></li><li><a href="https://developer.hashicorp.com/terraform/install">Terraform CLI</a></li><li><a href="https://nodejs.org/en/download/current">Node.js</a></li><li><a href="https://www.python.org/downloads/">Python</a></li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SvxqN_yNwc2lPqMxX86rlw.jpeg" /><figcaption>Using Claud Code within VS Code IDE</figcaption></figure><p>When I asked <a href="https://claude.com/product/claude-code">Claude Code</a> to deploy the application on AWS, it began with reading the cached credentials from the AWS CLI. This gets set when calling the below command:</p><pre># Supply<br># AWS Access Key ID<br># AWS Secret Access Key<br># Default region name<br># Default output format<br>aws configure</pre><p>Then it prompted me in the form of a numbered list on various options and asked me, “How do you want to deploy this app to AWS?”:</p><ul><li>ECS Fargate + S3/CloudFront</li><li>Single EC2 instance</li><li>Lambda + API Gateway</li><li>App Runner + S3/CloudFront</li></ul><p>Having some prior experience with API Gateway, I chose that option. Based on my choice, <a href="https://claude.com/product/claude-code">Claude Code</a> chose this tech stack for me. I had to explicitly prompt <a href="https://claude.com/product/claude-code">Claude Code</a> to use Terraform instead of Cloud Formation, since I was more familiar with it:</p><p><strong>Frontend</strong></p><p>React 18 (SPA) with Vite 8 as the build tool/dev server<br> — <a href="http://twitter.com/vitejs/plugin-react">@vitejs/plugin-react</a> 6 (bumped from 4 for Vite 8 compatibility)<br> — Tailwind CSS 3 (+ PostCSS, Autoprefixer) for styling<br> — Plain fetch for API calls (computes x-amz-content-sha256 via the browser Web Crypto API for OAC signing)<br> — Node 24 / npm 11 toolchain</p><p><strong>Backend</strong></p><p>Python 3.13 (Lambda runtime; 3.14 locally)<br> — FastAPI + Pydantic v2 (request validation, /api/evaluate route)<br> — Anthropic Python SDK calling Claude claude-sonnet-4–6 (with prompt caching)<br> — Mangum — ASGI→Lambda adapter<br> — Uvicorn for local dev; python-dotenv for local env loading</p><p><strong>Infrastructure / Deployment</strong> (AWS, us-east-1)</p><p>Terraform (v1.5.7, AWS provider ~5.0) — all infra as code<br> — AWS Lambda (zip package) behind a Lambda Function URL (AWS_IAM auth)<br> — Amazon CloudFront — CDN fronting both origins, with Origin Access Control (OAC) SigV4 signing<br> — Amazon S3 — private bucket for the static frontend build<br> — CloudWatch Logs for the Lambda<br> — Docker — build sandbox only, to produce Linux-native dependency wheels for Lambda</p><h3>Experience</h3><p>It was very entertaining to see in the console how <a href="https://claude.com/product/claude-code">Claude Code</a> was logging its reasoning and fixing issues. I used Opus 4.8. At one point, it determined that re-architecting the application was required, since the call to the Anthropic API from the application took at times more time than the timeout setting on the AWS API GW. It switched to using Lambda function URL. The final architecture looks like this:</p><figure><img alt="High level architecture of the Idea Evaluator application" src="https://cdn-images-1.medium.com/max/1024/1*9EzR8BBTdO6tz67ONGbBcw.png" /></figure><p>The Lambda function has access to the Anthropic API key through its environment variable, where the value is injected through a terraform variable at the time of the Lambda function deployment. The landing page of the application is served via a CloudFront distribution. In the AWS console, it is setup like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*d-4U6fcY7IYf_P28YIDZow.jpeg" /><figcaption>CloudFront distribution setup</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hzCP5JRPTHmJUPZh4LrKSg.jpeg" /><figcaption>Origin configuration</figcaption></figure><p><a href="https://aws.amazon.com/cloudfront/">CloudFront</a> acts as the facade to both the static content served off an S3 bucket and serverless API served via a Lambda function. No external party can directly access the S3 bucket nor invoke the Lambda function.</p><p>It took around 38 minutes of <a href="https://claude.com/product/claude-code">Claude Code</a> babble and me continuously hitting the Yes option, until I could finally get a working application. If you’d like to see the source code, check out: <a href="https://github.com/decliffy/idea-evaluator-tool">https://github.com/decliffy/idea-evaluator-tool</a></p><h3>Demos</h3><p>Here I’d like to present pre-recorded demos and saved outputs to showcase the tool in action. I had 3 test runs of the tool.</p><p><strong>Run # 1 Input</strong>: “An AI-powered website portal where gig artists can upload video samples of their performances. A writeup would be created automatically and include details such as the genre of the music covered, songs played, AI critique of the music. After meeting certain pre-determined criteria, the gig artist would be listed on the portal along with all the above details where potential clients could browse and enlist the artist for paid performances. The portal makes money from registration fees.”</p><iframe src="https://cdn.embedly.com/widgets/media.html?url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DkjMDdtzqh0s&amp;type=text%2Fhtml&amp;schema=google&amp;display_name=YouTube&amp;src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FkjMDdtzqh0s" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/1b4eb8b6b9fd952275cecaed901c9f96/href">https://medium.com/media/1b4eb8b6b9fd952275cecaed901c9f96/href</a></iframe><p><strong>Output</strong>: <a href="https://github.com/decliffy/idea-evaluator-tool/blob/main/demo/Idea%20Evaluator%20Output%20-%20Ex01.pdf">https://github.com/decliffy/idea-evaluator-tool/blob/main/demo/Idea%20Evaluator%20Output%20-%20Ex01.pdf</a></p><p><strong>Run # 2 Input</strong>: “An AI‑driven website portal where independent home‑cleaning professionals can create service profiles by uploading short video demonstrations of their cleaning techniques, tools used, and before‑and‑after results. The system automatically generates a detailed write‑up for each cleaner, including the types of cleaning they specialize in (deep cleaning, kitchen sanitization, upholstery care, bathroom detailing, etc.), estimated service duration, cleanliness quality score, and an AI‑based evaluation of technique, efficiency, and safety practices.</p><p>Once a cleaner meets predefined quality benchmarks — such as video clarity, demonstrated skill level, and safety compliance — the platform publishes their profile publicly. Potential clients can browse cleaners by service type, rating, location, and price range, then book them for paid home‑cleaning jobs.</p><p>The portal earns revenue through cleaner registration fees and optional premium placement upgrades.”</p><iframe src="https://cdn.embedly.com/widgets/media.html?url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DOuAePG-GZ-I&amp;type=text%2Fhtml&amp;schema=google&amp;display_name=YouTube&amp;src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FOuAePG-GZ-I" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/0c56e5c48a2da772e223e8a1aa975f5c/href">https://medium.com/media/0c56e5c48a2da772e223e8a1aa975f5c/href</a></iframe><p><strong>Output</strong>: <a href="https://github.com/decliffy/idea-evaluator-tool/blob/main/demo/Idea%20Evaluator%20Output%20-%20Ex02.pdf">https://github.com/decliffy/idea-evaluator-tool/blob/main/demo/Idea%20Evaluator%20Output%20-%20Ex02.pdf</a></p><p><strong>Run # 3 Input</strong>: “AI‑driven website portal where small‑scale home food vendors can create product profiles by uploading short videos showcasing how their dishes are prepared, the ingredients used, hygiene practices followed, and the final presentation of the food. The system automatically generates a detailed write‑up for each vendor, including cuisine type (Indian, continental, fusion, regional specialties), dish categories (meals, snacks, desserts, beverages), ingredient lists, estimated shelf life, portion size, and an AI‑based evaluation of taste indicators, hygiene standards, and preparation quality.</p><p>Once a vendor meets predefined quality benchmarks — such as hygiene compliance, ingredient transparency, and video authenticity — the platform publishes their listing publicly. Customers can browse home‑prepared food options by cuisine, dietary preference, location, price range, and vendor rating, then place orders for home delivery.</p><p>The portal earns revenue through vendor registration fees and optional promotional upgrades for better visibility.”</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FARjGZjm-S8I%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DARjGZjm-S8I&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FARjGZjm-S8I%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/6212fbe6d57ff27219f14ac2c4e4786d/href">https://medium.com/media/6212fbe6d57ff27219f14ac2c4e4786d/href</a></iframe><p><strong>Output</strong>: <a href="https://github.com/decliffy/idea-evaluator-tool/blob/main/demo/Idea%20Evaluator%20Output%20-%20Ex03.pdf">https://github.com/decliffy/idea-evaluator-tool/blob/main/demo/Idea%20Evaluator%20Output%20-%20Ex03.pdf</a></p><h3>Conclusion</h3><p>The experience building and deploying this application with <a href="https://claude.com/product/claude-code">Claude Code</a> is unlike any other experience I have had in the past. The ability to build something that works not just on the local machine but on a public cloud is fantastic. Instead of AWS, I could have chosen Azure or Google Cloud, and I could expect to still be successful getting the application running on those platforms. I haven’t written a single line of code, yet my past experience enabled me to understand what <a href="https://claude.com/product/claude-code">Claude Code</a> is doing and to also guide it in the right direction. The overall effort to get to this point has been just under 2 hours. Mission accomplished and on to greater things with <a href="https://claude.com/product/claude-code">Claude Code</a> !</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*NS50RJL1r0WLbxr3" /><figcaption>Photo by <a href="https://unsplash.com/@silverkblack?utm_source=medium&amp;utm_medium=referral">Vitaly Gariev</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dc4c7ce6b59a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[My First Steps With Claude Code]]></title>
            <link>https://cliffordedsouza.medium.com/my-first-steps-with-claude-code-7755caecb06f?source=rss-b2feb9ac956------2</link>
            <guid isPermaLink="false">https://medium.com/p/7755caecb06f</guid>
            <category><![CDATA[anthropic-claude]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[claude-code]]></category>
            <dc:creator><![CDATA[Clifford E. D'Souza]]></dc:creator>
            <pubDate>Fri, 08 May 2026 18:29:39 GMT</pubDate>
            <atom:updated>2026-05-09T07:02:09.116Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*qTKNyAvUtTX5xNND" /><figcaption>Photo by <a href="https://unsplash.com/@hdbernd?utm_source=medium&amp;utm_medium=referral">Bernd 📷 Dittrich</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>As a first time user of Claude Code I knew that I had to cross the chasm by customarily building a working application entirely using an AI coding agent. I would like to share my journey in building a <strong>software effort estimator</strong>. I started from a clean slate and ended up with working software!</p><p>Software development teams don’t struggle because work is hard — they struggle because <strong>not all work is created equal</strong>, yet we treat effort estimation as if it is. Enhancing a fragile, production‑critical component is nothing like building a brand‑new greenfield feature, but our planning rituals blur these distinctions.</p><p>To make matters worse, estimating effort still depends on pulling senior engineers into long, interruption‑heavy conversations just to arrive at rough T‑shirt sizes. This process can drag on for days, produces wildly inconsistent outputs, and often reflects the biases or heuristics of whoever happened to be in the room.</p><p>The result is a planning function that is slow, subjective, and unreliable — precisely when organizations need clarity, speed, and predictability the most.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*-LpI-mMdbZT-Al_V" /><figcaption>Photo by <a href="https://unsplash.com/@dominapetric?utm_source=medium&amp;utm_medium=referral">Domina Petric</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>This made me think of a use case for leveraging the power of Large Language Models (LLMs) to help with these problems. And I knew that I needed to take a new path towards building a solution, one that involved leveraging agentic AI. I didn’t have a clear idea on how I would need to proceed, that didn’t stop me though.</p><p>I knew that I had to setup a workflow, starting from specifications that would ultimately transform into working software. The break through for me came when I discovered <a href="https://github.com/github/spec-kit">GitHub Spec Kit</a> (<a href="https://github.com/github/spec-kit">https://github.com/github/spec-kit</a>). In order to get started, I first installed <a href="https://claude.com/product/claude-code">Claude Code</a> (<a href="https://claude.com/product/claude-code">https://claude.com/product/claude-code</a>) which is an AI coding agent. This requires a subscription. I also needed a CLI tool to interact with my GitHub repositories. I installed <a href="https://cli.github.com/">GitHub CLI</a> (<a href="https://cli.github.com/">https://cli.github.com/</a>) for that. One last thing was to buy some credits on the <a href="https://platform.claude.com/">Anthropic platform</a> (<a href="https://platform.claude.com/">https://platform.claude.com/</a>). This is needed to generate an Anthropic API key which would be required later. I also needed the <a href="https://docs.astral.sh/uv/">uv tool</a> (<a href="https://docs.astral.sh/uv/">https://docs.astral.sh/uv/</a>) which I installed locally. These were all that I required to get started to achieve my goal.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*s-mwuLU85XWsGjs6r9SdFg.png" /><figcaption>Post GitHub Spec Kit completion message</figcaption></figure><p>Still not knowing where I would end up, I took the dive with the first step. I created an empty folder and simply named it <strong>effort-estimator</strong>. Then, following the GitHub spec kit documentation, I ran the following command from inside the folder:</p><pre>uvx - from git+https://github.com/github/spec-kit.git@v0.8.7 specify init - here - integration claude</pre><p>The next easy step for me was to start up Claude Code by typing <strong>claude</strong> in the command line and hitting enter. In the Claude Code console, I verified GitHub Spec Kit skills were installed by keying in /speckit-<br>Yes, the skills were installed, and I could see /speckit-specify, /speckit-plan, /speckit-tasks. This is a good start I thought at that time. My Eureka moment of sorts :)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pBpDfTgXRPBDY256yVjuRQ.png" /><figcaption>Validate GitHub Spec Kit skills in Claude</figcaption></figure><p>The GitHub Spec Kit documentation stated that I had to run the below sequence. I just went ahead granting permission to the agent by choosing <strong>Yes</strong> all the time.</p><pre>/speckit-constitution Create principles focused on code quality, testing standards, user experience consistency, and performance requirements<br><br>/speckit-specify Build an application that can help me estimate T-shirt size effort for software development tasks<br><br>/speckit-plan<br><br>/speckit-tasks<br><br>/speckit-implement</pre><p>With a few iterations and plenty of tokens expended, I could get the final output committed to my <a href="https://github.com/decliffy/effort-estimator">GitHub public repository</a> (<a href="https://github.com/decliffy/effort-estimator">https://github.com/decliffy/effort-estimator</a>)</p><p>The results look good. Leveraging the power of Anthropic LLM models, for a set of tasks supplied as input, I could get the following output:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1K5P0vL5c0ZVa27LunAg2Q.png" /><figcaption>Output from execution of the program</figcaption></figure><p>I could also run the tests and see that they all pass.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*n3a1m88uwixg7pp6te9JaA.png" /><figcaption>Test execution</figcaption></figure><p>This <a href="https://github.com/decliffy/effort-estimator">GitHub repository</a> can be cloned and checked out, and with an Anthropic API key supplied in an environment variable, the application can be run to provide T-shirt estimates on any system. I did not have to write a single line of code. And, as I got a working piece of software at the end, my confidence grew. If you are someone looking to dive in to agentic AI, I invite you to try out the same steps I followed for building your own idea.</p><p>The application that I built supports a command line interface. I would be keen to know how it went if you decided to clone my repository and extend what I already built. Perhaps, add a browser based UI, or create a pipeline to deploy the application to a cloud provider such as AWS or Azure. If you do take up the challenge, one thing is sure, you’d also get on the track to an AI assisted journey!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*D-XggMT3Nyj4juWZ" /><figcaption>Photo by <a href="https://unsplash.com/@gribgrab?utm_source=medium&amp;utm_medium=referral">Gribgrab</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7755caecb06f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Age of Applied Intelligence: Transforming Problems Into Products with AI AgentsProblem…]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://cliffordedsouza.medium.com/the-age-of-applied-intelligence-transforming-problems-into-products-with-ai-agentsproblem-a278e5a28c8c?source=rss-b2feb9ac956------2"><img src="https://cdn-images-1.medium.com/max/2600/0*rjaC1z0kfF32tiUf" width="3840"></a></p><p class="medium-feed-snippet">Background</p><p class="medium-feed-link"><a href="https://cliffordedsouza.medium.com/the-age-of-applied-intelligence-transforming-problems-into-products-with-ai-agentsproblem-a278e5a28c8c?source=rss-b2feb9ac956------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://cliffordedsouza.medium.com/the-age-of-applied-intelligence-transforming-problems-into-products-with-ai-agentsproblem-a278e5a28c8c?source=rss-b2feb9ac956------2</link>
            <guid isPermaLink="false">https://medium.com/p/a278e5a28c8c</guid>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[scalable-automation]]></category>
            <category><![CDATA[innovation-execution-gap]]></category>
            <category><![CDATA[agentic-workflow]]></category>
            <category><![CDATA[ai-native-product]]></category>
            <dc:creator><![CDATA[Clifford E. D'Souza]]></dc:creator>
            <pubDate>Sun, 08 Mar 2026 06:22:57 GMT</pubDate>
            <atom:updated>2026-03-08T06:22:57.985Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Setting up Autogen Notebook]]></title>
            <link>https://cliffordedsouza.medium.com/setting-up-autogen-notebook-6e64338848ae?source=rss-b2feb9ac956------2</link>
            <guid isPermaLink="false">https://medium.com/p/6e64338848ae</guid>
            <category><![CDATA[autogen]]></category>
            <category><![CDATA[python]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[llm]]></category>
            <dc:creator><![CDATA[Clifford E. D'Souza]]></dc:creator>
            <pubDate>Mon, 31 Mar 2025 17:07:32 GMT</pubDate>
            <atom:updated>2025-03-31T17:07:32.333Z</atom:updated>
            <content:encoded><![CDATA[<p>Wanted to share how I setup one of the Jupyter notebooks as available in the online course <a href="https://www.deeplearning.ai/short-courses/ai-agentic-design-patterns-with-autogen/">AI Agentic Design Patterns with AutoGen</a> from DeepLearning.ai locally on my PC.</p><p>First step was to login to the <a href="https://platform.openai.com/">Open AI API platform</a> for the purpose of generating an API key and acquiring Credits. Once logged in the Open AI API platform, under Settings-&gt;API keys, I created a new API key and stored it securely. Credits can be purchased from the Billing section.</p><h3><strong>Setting up local</strong></h3><p>Here I followed the recommended approach, which is to create a python virtual environment, activate it and then start installing the required modules. I ran these commands in a working directory.</p><pre>python -m venv myenv<br><br>source myenv/bin/activate<br><br>pip install jupyter</pre><p>In the same folder, I downloaded the first notebook example from the above-mentioned course along with the dependent files. My local folder structure looks as below.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_Kw7Q6QFZxTB60cxNVyGlQ.jpeg" /><figcaption>Local folder structure</figcaption></figure><p>I ran the code below for installing the dependencies. One of the key dependencies is pyautogen</p><pre>pip install -r requirements.txt</pre><p>Now, just needed a few more steps, namely, to set an environment variable and start up the Jupyter notebook server, accomplished through the steps below.</p><pre>export OPENAI_API_KEY=&lt;&lt;My Open API Key&gt;&gt;<br><br>jupyter notebook</pre><p>I pushed the notebook to my <a href="https://github.com/decliffy/autogen/blob/main/notebooks/L1_Multi-Agent_Conversation_and_Stand-up_Comedy.ipynb">github repository</a></p><p>All the cells execute correctly locally, just as in the online course hosted notebook. The local environment is therefore now setup to experiment with agentic AI on my PC.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EIYS4H0m3Z-1HzDNfIRy0w.png" /><figcaption>Example Jupyter notebook running locally with connectivity to Open AI API</figcaption></figure><p>Other notebooks could also be downloaded at this point to allow experimentation outside of the boundary of the online course. Useful references to leverage include the official <a href="https://github.com/microsoft/autogen">Autogen GitHub repository</a> and other Autogen repositories available online.</p><p>My goal was that I wanted to setup the notebook examples locally and was successful in doing so. This also sets the stage to experiment further with Autogen locally on my PC.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6e64338848ae" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[ChatGPT for Developer Productivity]]></title>
            <link>https://cliffordedsouza.medium.com/chatgpt-for-developer-productivity-9dd97001c447?source=rss-b2feb9ac956------2</link>
            <guid isPermaLink="false">https://medium.com/p/9dd97001c447</guid>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[testing]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[developer]]></category>
            <dc:creator><![CDATA[Clifford E. D'Souza]]></dc:creator>
            <pubDate>Mon, 28 Aug 2023 06:15:02 GMT</pubDate>
            <atom:updated>2023-08-28T06:27:09.943Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9RUrfC7Av1uUujMf9a96RA.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@ilgmyzin?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">ilgmyzin</a> on <a href="https://unsplash.com/photos/agFmImWyPso?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Unsplash</a></figcaption></figure><p><strong>Introduction</strong></p><p>In today’s world of software engineering, crafting tests has become a vital companion to writing code. This is due to the rapid pace at which code evolves, especially within agile development methods. With practices like shared code ownership and continuous integration, automated test cases play a crucial role. They not only ensure that all existing code functions correctly, aligning with the developers’ intentions, but also serve as guides when passing on the code to fellow developers for future work. The dual responsibilities of creating functional code and its corresponding tests must match the speed of innovation and the accelerating release cycles.</p><p>However, the race to decrease lead time — measured from when a task is selected for work to its actual release — cannot be won solely by churning out more code at a faster rate. The solution lies in automating repetitive tasks. This smarter approach paves the way for efficiency, allowing more time and effort to be directed towards innovation and other value-added activities.</p><p><strong>Taking a Deeper look</strong></p><p>When a user story is picked up by a developer from the sprint backlog, the first thing looked at from an analysis perspective is the Acceptance criteria. Code for the feature should be written in such a way that the acceptance criteria are met all the time. Frameworks such as Cucumber can help to automate Behaviour Driven Development (BDD) code, which helps to assert that the newly written code is meeting the acceptance criteria. This is all good. However, starting to write these tests from a blank slate requires effort and sometimes includes that effort of finding similar code to customise for the current task. What if we could search for a template to get started and move ahead more quickly?</p><p>Let’s say the software engineer spends 4 or 5 days writing the code and the acceptance tests pass. Now is the time to check-in the code. However, it needs to go through a PR (Pull request) that triggers a step to peer review the code before the code could be merged into the develop branch. Everyone’s usually busy working on their own code within a scrum sprint, and certainly, an in-depth review of a PR can take a significant effort on the part of both the code reviewers and the code author. Effort needs to be spent on understanding the requirement, the context, the assumptions and determine if there are any logical errors with the code that if not rectified could lead to a potential issue in production. The coding standards also need to be adhered to as well as ascertaining the fact that the code is professionally written. This includes looking out for how well the code is written, that it can handle the edge cases, and generally that the new code doesn’t degrade the overall standards, performance, security, and quality of the existing code. This effort, while beneficial, often does not get tracked in terms of effort expended. Most of the time, developers want to get done with this step as soon as possible, and considering the time constraints, are sometimes willing to lower the bar a little.</p><p>Unless following an approach of Test Driven Development (TDD), tests tend to be written after the code has been written. Often the bare minimum unit tests are written to sustain a minimum level of test coverage. However, the key here is to cover all the flows in the code and validate how the code handles all the potential input parameter values. Broader the gap here, more the chances of getting bugs in the code. Testing all possible scenarios can be hard work, and when time is a constraint, some of the tests could get skipped.</p><p><strong>Listing the Problems</strong></p><p>We have three underlying problems identified from the preceding section:</p><ul><li>There needs to be better ways to quickly get started with writing automated BDD test cases to meet the acceptance criteria.</li><li>There needs to be a better way of making peer code reviews more effective.</li><li>Test cases need to be auto-generated and in manner that reveals software bugs or prompt the software engineer for a conversation with the product team to clarify scenarios.</li></ul><p>Let’s explore how chatGPT could provide solutions to these problems.</p><p><strong>Automated BDD testing</strong></p><p>We can go prompt by prompt at <a href="http://chat.openai.com/">chat.openai.com</a> with a logical sequence of prompts, making sure we are getting the expected response from chatGPT.</p><p>First, we start with this prompt:</p><ul><li>Let us assume we have the following acceptance criteria for a user story:</li></ul><p>With nothing else provided, this prompts chatGPT to ask for additional inputs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/776/1*1Iz1jvxjDP8alo0TVzmiFw.png" /></figure><p>We then supply acceptance criteria and then get a further result from chat GPT:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/691/1*hlDatK2vaXY6zNllIZWT8w.png" /></figure><p>Now, let’s go for the code generation section. ChatGPT generates the feature file, which specifies how the system under development is expected to behave under varying scenarios.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/671/1*WhhB1zxo_AwW-WRLJNwlMg.png" /></figure><p>Full listing here:</p><pre>Feature: Display Account Information<br><br>  Scenario: Display statement balance upon authentication<br>    Given the user is authenticated<br>    When the user views their account information<br>    Then the statement balance should be displayed<br><br>  Scenario: Show minimum payment due and payment due date<br>    Given the user is authenticated<br>    When the user views their account information<br>    Then the minimum payment due and payment due date should be displayed<br><br>  Scenario: Show error message if service not responding or timeout<br>    Given the user is authenticated<br>    When the account information service is not responding<br>    Then an error message indicating a service issue should be displayed</pre><p>This continues further into the output of the steps definitions via Cucumber step definition code in Java.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/625/1*1cB_FhhaZ94u4jxQKdMwow.png" /></figure><p>Full Source code:</p><pre>import cucumber.api.java.en.Given;<br>import cucumber.api.java.en.When;<br>import cucumber.api.java.en.Then;<br><br>public class AccountInformationSteps {<br><br>    @Given(&quot;the user is authenticated&quot;)<br>    public void userIsAuthenticated() {<br>        // Code to perform user authentication<br>    }<br><br>    @When(&quot;the user views their account information&quot;)<br>    public void userViewsAccountInformation() {<br>        // Code to navigate to account information page<br>    }<br><br>    @Then(&quot;the statement balance should be displayed&quot;)<br>    public void statementBalanceShouldBeDisplayed() {<br>        // Code to verify statement balance is displayed<br>    }<br><br>    @Then(&quot;the minimum payment due and payment due date should be displayed&quot;)<br>    public void minimumPaymentAndDueDateShouldBeDisplayed() {<br>        // Code to verify minimum payment due and due date are displayed<br>    }<br><br>    @When(&quot;the account information service is not responding&quot;)<br>    public void accountServiceNotResponding() {<br>        // Code to simulate service not responding<br>    }<br><br>    @Then(&quot;an error message indicating a service issue should be displayed&quot;)<br>    public void errorMessageForServiceIssueShouldBeDisplayed() {<br>        // Code to verify error message is displayed<br>    }<br>}</pre><p>As the chatGPT output states, and as is pretty evident:</p><ul><li>The actual implementation of these steps will depend on your application’s code structure and the tools you’re using for testing. Make sure to adapt the step definitions to match your application’s behavior and the testing libraries you’re using with Cucumber.</li></ul><p>Additionally, you’ll need to set up the Cucumber framework and the necessary dependencies in your project to be able to run these BDD tests. The code provided here is a basic example and may need further customization to match your specific testing environment.</p><p><strong>Peer code review with ChatGPT</strong></p><p>While not intended to replace a human code reviewer, we can already see the potential of helping a reviewer out. Here is an example, and the potential opportunity for educating developers on how to perform an effective code review:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rJcoZy4fGq9z-4y31CigcQ.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/631/1*oQ5HmXHPN3g_3owKwTVY7w.png" /></figure><p>Isn’t this helpful? The basics of a good code review are suggested and could help a code reviewer with a context specific checklist of things to check in the newly written code. This saves time. ChatGPT also goes on to further suggest how the code could be improved.</p><p><strong>Generate Test Cases</strong></p><p>Whether it is Test Driven Development (TDD) or simply writing Unit tests, one of the initial tasks is to produce a list of test cases from the feature requirements. The tests can then be written starting from these lists of tests. Here’s an example of generation of an initial set of test cases. We begin by simply prompting chatGPT with the prompt below:</p><ul><li>Generate an exhaustive list of test cases</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/824/1*GXiKotndSv4Pf22kJfO8LQ.png" /></figure><p>Here’s the full output from chatGPT. This list seems exhaustive and includes Non-functional requirement test cases as well!</p><p><strong>Thumbs-up Voting Test Cases</strong>:<br> — Verify that users can see a “thumbs-up” icon associated with each feature request.<br> — Test if users can click the “thumbs-up” icon to vote for a feature request.<br> — Check if the vote count increases when a user clicks the “thumbs-up” icon.<br> — Ensure that users can click the “thumbs-up” icon only once per feature request.<br> — Test if the “thumbs-up” icon changes visually (e.g., color) after a user has voted.<br> — Verify that users cannot vote on their own feature requests.<br> — Test if the vote count updates in real-time for all users viewing the feature request.<br> — Check if users can change their vote from a “thumbs-up” to “thumbs-down” or vice versa.<br> — Ensure that administrators or moderators have the ability to remove votes if needed.<br> — Test the behavior of the “thumbs-up” feature on various devices and browsers.</p><p><strong>Commenting Test Cases</strong>:<br> — Verify that users can see a comment section associated with each feature request.<br> — Test if users can type and submit a comment on a feature request.<br> — Check if submitted comments are displayed correctly with the username, timestamp, and content.<br> — Ensure that users can edit their own comments after submission.<br> — Test if users can delete their own comments from a feature request.<br> — Verify that users cannot edit or delete comments from other users.<br> — Test if the comment section supports formatting (e.g., bold, italics, bullet points).<br> — Check if URLs posted in comments are automatically hyperlinked.<br> — Ensure that long comments are displayed properly without layout issues.<br> — Test the behavior of the commenting feature on various devices and browsers.</p><p><strong>Integration Test Cases</strong>:<br> — Verify that the total vote count accurately reflects the sum of all “thumbs-up” votes.<br> — Test if the vote count and comments update properly when the page is refreshed.<br> — Check if users can see the highest voted feature requests listed at the top.<br> — Ensure that comments and votes are properly synchronized between different clients.<br> — Test the performance of the feature when there are a large number of votes and comments.</p><p><strong>Security and Permissions Test Cases</strong>:<br> — Verify that only authenticated users can vote or comment.<br> — Test if unauthorized users are restricted from voting or commenting.<br> — Check if the feature request owner has the ability to moderate comments (edit or delete).<br> — Ensure that only administrators or authorized users can remove comments.<br> — Test for potential security vulnerabilities, such as injecting scripts into comments.</p><p><strong>Usability and User Experience Test Cases</strong>:<br> — Verify that the “thumbs-up” icon and comment section are intuitively located.<br> — Test if users receive clear feedback when they vote or submit a comment.<br> — Check if error messages are displayed appropriately for failed actions (e.g., voting without authentication).<br> — Ensure that the user’s own comments and votes are visually distinct from others’.<br> — Test the responsiveness and load times of the feature under various network conditions.</p><p>ChatGPT adds a disclaimer at the end of the output: Remember that this list is extensive and might need to be tailored to your specific application and requirements.</p><p><strong>Conclusion</strong></p><p>Going beyond the IDE, chatGPT could be leveraged as an invaluable additional tool in the software engineer&#39;s toolkit for assisting in yet more effective and efficient development practices.</p><p>As we’ve seen here, chatGPT can help with boosting the pace of software development by defining the initial structure of test cases that the software engineer can focus on writing the code for, one by one. The time taken to think of different test cases could be reduced significantly, with the initial set of test cases being able to be easily generated with chatGPT.</p><p>Code reviews being assisted with prompts can make the job of the code reviewer easier than before. Developers don’t need to wait till the PR stage to get their code reviewed. As a personal productivity tool, the first pass of code reviews could be run by chatGPT even as the code is written incrementally.</p><p>Finally, with a ready set of test cases generated by chatGPT, software engineers could get started writing unit tests and uncover scenarios that could otherwise have led to bugs.</p><p>We could expect further improvements in the domain of Generative AI and coding. It is 2023, and the field already looks promising using ChatGPT.</p><p><strong>Other Interesting Reads</strong></p><ul><li><a href="https://www.codium.ai/blog/chatgpt-for-automated-testing-examples-and-best-practices/">ChatGPT for Automated Testing: Examples and Best Practices</a></li><li><a href="https://smartbear.com/blog/how-to-use-chatgpt-in-your-automated-tests/">How to Use ChatGPT in Your Automated Tests</a></li><li><a href="https://www.qable.io/chatgpt-in-testing/">ChatGPT in Testing- Importance, Uses, Limitations for QA</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9dd97001c447" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Unlocking Business Value: The Importance of Considering Business Value when Starting an Applied…]]></title>
            <link>https://cliffordedsouza.medium.com/unlocking-business-value-the-importance-of-considering-business-value-when-starting-an-applied-960a03f7771e?source=rss-b2feb9ac956------2</link>
            <guid isPermaLink="false">https://medium.com/p/960a03f7771e</guid>
            <category><![CDATA[applied-ai]]></category>
            <category><![CDATA[business-value]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Clifford E. D'Souza]]></dc:creator>
            <pubDate>Thu, 29 Jun 2023 14:46:59 GMT</pubDate>
            <atom:updated>2023-06-29T14:46:59.657Z</atom:updated>
            <content:encoded><![CDATA[<h3>Unlocking Business Value: The Importance of Considering Business Value when Starting an Applied Artificial Intelligence Product</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dTRUucJVU4ww4NEGUs47TA.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@mikofilm?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Mike Kononov</a> on <a href="https://unsplash.com/s/photos/Business-value-and-artificial-intelligence?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Unsplash</a></figcaption></figure><h3>Introduction</h3><p>Artificial Intelligence (AI) has rapidly emerged as a transformative technology, revolutionizing various industries and business processes. From healthcare to finance, AI-powered solutions have the potential to enhance efficiency, automate tasks, and generate significant business value. However, before diving into the development of an applied AI product, it is essential to consider the business value it can deliver. By aligning AI initiatives with strategic goals and understanding the potential return on investment, businesses can maximize the benefits of their AI endeavors.</p><h3>Understanding Business Goals</h3><p>Before embarking on an applied AI project, it is crucial to have a clear understanding of your organization’s business goals. AI should be seen as a means to achieve those goals, rather than an end in itself. Determine the specific challenges or opportunities you aim to address using AI. Whether it’s improving customer service, optimizing supply chains, or enhancing product recommendations, aligning AI initiatives with business objectives is fundamental to realizing its value.</p><h3>Identifying High-Impact Use Cases</h3><p>To maximize business value, it is important to identify and prioritize high-impact use cases where AI can make a significant difference. Engage with stakeholders from different departments and functions to gather insights on pain points, inefficiencies, or areas that can benefit from intelligent automation. Evaluate the potential impact of AI solutions on key performance indicators, such as cost reduction, revenue growth, customer satisfaction, or operational efficiency. Focusing on high-value use cases ensures that your AI investment yields tangible results and generates a positive return.</p><h3>Quantifying Return on Investment (ROI)</h3><p>To make informed decisions about AI investments, it is essential to quantify the potential return on investment. Assess the costs associated with developing and deploying an AI solution, including data acquisition, infrastructure, talent, and ongoing maintenance. Simultaneously, estimate the potential benefits in financial terms, such as increased revenue, cost savings, or improved productivity. By comparing costs and benefits, you can determine the viability and profitability of your AI initiative, enabling more effective resource allocation.</p><h3>Consider Ethical and Legal Implications</h3><p>While evaluating the business value of an AI product, it is essential to consider ethical and legal implications. AI systems must be designed to uphold privacy, security, and fairness. Assess the potential risks associated with data privacy, bias, and algorithmic transparency. Compliance with relevant regulations, such as GDPR or HIPAA, should be a top priority. By ensuring ethical and legal compliance, you protect your business reputation and avoid potential pitfalls that may erode the anticipated business value of your AI product.</p><h3>Building a Scalable and Sustainable Model</h3><p>To drive long-term business value, it is crucial to build a scalable and sustainable AI model. Consider factors such as data availability, system integration, and adaptability to changing business requirements. Developing a robust infrastructure capable of handling large-scale data and accommodating future growth is vital. Additionally, focus on developing AI models that can evolve and learn from new data, ensuring their relevance and effectiveness over time.</p><h3>Conclusion</h3><p>When embarking on an applied AI project, it is essential to think about the business value it can deliver. Aligning AI initiatives with business goals, identifying high-impact use cases, quantifying return on investment, considering ethical and legal implications, and building scalable models are all critical steps in maximizing the value of AI for your organization. By thinking strategically about AI and its potential impact, businesses can unlock new opportunities, gain a competitive edge, and drive sustainable growth in the era of artificial intelligence.</p><h3>Additional Reading</h3><p><a href="https://sloanreview.mit.edu/article/achieving-return-on-ai-projects/">Achieving Return on AI Projects (mit.edu)</a></p><p><a href="https://www.ibm.com/downloads/cas/ZENVBND4">The business value of AI (ibm.com)</a></p><p><a href="https://www.pragmaticinstitute.com/resources/articles/data/making-the-leap-from-ai-investments-to-business-results/">Making the Leap from AI Investments to Business Results — Pragmatic Institute Resources</a></p><p><a href="https://www2.deloitte.com/us/en/insights/focus/tech-trends/2019/driving-ai-potential-organizations.html">Driving AI’s potential in organizations | Deloitte Insights</a></p><p><a href="https://www.techtarget.com/searchenterpriseai/feature/Learn-the-business-value-of-AIs-various-techniques">Learn the business value of AI’s various techniques | TechTarget</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=960a03f7771e" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>