<?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 ConcertIDC Engineering on Medium]]></title>
        <description><![CDATA[Stories by ConcertIDC Engineering on Medium]]></description>
        <link>https://medium.com/@concertidc?source=rss-86328d0c504f------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*ZTPjrBCSNkoKLqmxbt8eKA.png</url>
            <title>Stories by ConcertIDC Engineering on Medium</title>
            <link>https://medium.com/@concertidc?source=rss-86328d0c504f------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 14:49:31 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@concertidc/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[FRONTEND CHALLENGES FOR AI PRODUCTS]]></title>
            <link>https://concertidc.medium.com/frontend-challenges-for-ai-products-1d977f860771?source=rss-86328d0c504f------2</link>
            <guid isPermaLink="false">https://medium.com/p/1d977f860771</guid>
            <dc:creator><![CDATA[ConcertIDC Engineering]]></dc:creator>
            <pubDate>Tue, 27 Oct 2026 20:44:45 GMT</pubDate>
            <atom:updated>2026-10-27T20:44:45.760Z</atom:updated>
            <content:encoded><![CDATA[<p>AI-powered products introduce a different set of frontend engineering challenges from traditional applications. In a conventional interface, a frontend often requests structured data from an API and renders a predictable response. With AI, the content is generated dynamically: its length and format can vary, it may arrive incrementally, and the same input can produce different results. These characteristics affect rendering, state management, accessibility, security, error handling, and performance.</p><p><strong>Why frontend development is different for AI</strong></p><p>The central shift is from fixed data to generated content. A response might be a single sentence, a long explanation, a table, a code sample, or a mixture of formats. It may also stream to the browser piece by piece rather than arriving as one complete payload. The interface therefore needs to adapt while content is being produced — not just after the request finishes.</p><p>The following eight areas are useful design and implementation considerations for teams building conversational or generative AI experiences.</p><p><strong>1. Streaming and real-time responses</strong></p><p><strong>The challenge. </strong>AI responses may arrive token by token. Replacing the entire message on every update can cause flicker, waste rendering work, and make the page feel unstable. Scrolling also needs care: automatically moving to the bottom is helpful when a user is following the response, but disruptive when they have scrolled up to read earlier content.</p><p><strong>Practical approaches</strong></p><ul><li>Append incremental content instead of replacing the whole response.</li><li>Auto-scroll only when the user is already near the bottom.</li><li>Buffer and reconcile incomplete Markdown or code blocks before rendering them.</li><li>If a stream fails, show a clear retry or resume state rather than leaving a partial answer unexplained.</li></ul><p><strong>Key takeaway. </strong>Streaming changes the frontend from rendering a finished result to managing a result that is still being assembled.</p><p><strong>2. Unpredictable output length and format</strong></p><p><strong>The challenge. </strong>A generated answer can range from one line to several pages. It may include Markdown, code, tables, or plain text, and can contain long URLs or unbroken strings that stress a layout. Empty or truncated output must also be handled.</p><p><strong>Practical approaches</strong></p><ul><li>Use flexible containers instead of fixed-height response boxes.</li><li>Render Markdown through a suitable, maintained renderer with safe fallbacks.</li><li>Provide horizontal scrolling for wide code blocks and tables where appropriate.</li><li>Detect empty responses and present a helpful state.</li></ul><p><strong>Key takeaway. </strong>Unlike a typical endpoint with a known response schema, a generative interface must be prepared for multiple valid content shapes.</p><p><strong>3. Latency and loading states</strong></p><p><strong>The challenge. </strong>Model response times can range from under a second to many seconds. A blank screen makes the product appear broken, while a generic spinner may not communicate what is happening. Progress indicators should set expectations without implying a precise completion percentage when that information is unavailable.</p><p><strong>Practical approaches</strong></p><ul><li>Show a thinking or typing indicator as soon as generation begins.</li><li>Display partial content as soon as it is available.</li><li>Offer time expectations only when they can be estimated responsibly.</li><li>Make cancel or stop generation an easy-to-find action.</li></ul><p><strong>Key takeaway. </strong>Clear feedback reduces uncertainty and gives users control while they wait.</p><p><strong>4. Rendering rich and untrusted content safely</strong></p><p><strong>The challenge. </strong>Model output is still untrusted input. It may contain HTML-like text, malformed Markdown, links, image references, or code. A renderer must not turn generated content into an avoidable security or stability risk.</p><p><strong>Practical approaches</strong></p><ul><li>Sanitize content before displaying any HTML-capable output.</li><li>Use well-tested Markdown and code-rendering libraries.</li><li>Validate links and image URLs before presenting them as active content.</li><li>Provide fallback rendering for malformed Markdown or unsupported content.</li></ul><p><strong>Key takeaway. </strong>Safe rendering should be treated as a core requirement, not a finishing touch.</p><p><strong>5. Errors, refusals, and low-quality responses</strong></p><p><strong>The challenge. </strong>Not every unsuccessful AI interaction is the same. A request can time out, be refused, hit a rate limit, or return an answer that is irrelevant or poor in quality. A single generic error message hides these differences and gives users little guidance.</p><p><strong>Practical approaches</strong></p><ul><li>Create distinct, human-readable states for timeouts, refusals, quota limits, and other failures.</li><li>Keep “retry” and “regenerate” as separate actions when their intent differs.</li><li>When appropriate, help users improve a prompt by adding context or rephrasing rather than blindly repeating it.</li><li>Explain rate limits and quota restrictions clearly.</li></ul><p><strong>Key takeaway. </strong>Good error UX explains what happened and what the user can do next.</p><p><strong>6. State management for conversational UI</strong></p><p><strong>The challenge. </strong>Conversation interfaces manage more than a single form submission. History must stay synchronized with the backend, and users may edit earlier messages, regenerate answers, or send follow-ups while other requests are in progress. State may also need to survive refreshes or reconnects.</p><p><strong>Practical approaches</strong></p><ul><li>Represent the conversation as an ordered message list, with clear rules for updates.</li><li>Track loading and request status per message instead of relying on one global loading flag.</li><li>Model edits and conversation branches explicitly if the product supports them.</li><li>Persist local conversation state where appropriate and reconcile it with the backend after reconnecting.</li></ul><p><strong>Key takeaway. </strong>Per-message state makes concurrent actions easier to reason about and prevents one request from incorrectly controlling the entire interface.</p><p><strong>7. Accessibility and trust in AI output</strong></p><p><strong>The challenge. </strong>Streaming updates can overwhelm assistive technology if every token is announced. Users should also be able to identify AI-generated content and understand the basis or uncertainty of an answer when that information is available.</p><p><strong>Practical approaches</strong></p><ul><li>Use ARIA live regions carefully; avoid announcing every token as a separate update.</li><li>Label AI-generated content consistently.</li><li>Provide sources, citations, or confidence cues when they are available and meaningful.</li><li>Test real streaming responses with screen readers and other assistive technologies.</li></ul><p><strong>Key takeaway. </strong>Accessibility needs to be tested against the dynamic behavior of the product, not only against static screens.</p><p><strong>8. Performance with large payloads</strong></p><p><strong>The challenge. </strong>Long conversations and large responses can increase DOM size and memory use. Re-rendering an entire message list for every stream update can become expensive, while large code blocks and tables may slow down interaction.</p><p><strong>Practical approaches</strong></p><ul><li>Virtualize long message histories when appropriate.</li><li>Update the message currently streaming rather than re-rendering the full conversation.</li><li>Render heavy content lazily when it is not immediately needed.</li><li>Trim, paginate, or otherwise manage very long histories according to product requirements.</li></ul><p><strong>Key takeaway. </strong>Useful signals include time to first token, responsiveness during streaming, frame rate, and memory growth over a long session.</p><p><strong>Best practices to carry forward</strong></p><ul><li>Design for partial, streaming, and variable-length content from the beginning.</li><li>Treat every model response as untrusted input and render it safely.</li><li>Represent errors, refusals, and low-quality responses as distinct user experiences.</li><li>Track loading and request state per message rather than globally</li><li>Test accessibility and performance using real streaming responses and long conversations.</li></ul><p><strong>Questions for engineering teams</strong></p><ul><li>Which of these challenges have already appeared in our product?</li><li>Where would shared components or reusable patterns improve consistency?</li><li>What should we standardize for streaming, loading, errors, and regeneration?</li><li>How will we measure responsiveness and accessibility in realistic AI sessions?</li></ul><p><strong>Conclusion</strong></p><p>Building a strong frontend for AI is not simply a matter of placing a chat box around a model API. Generated content changes how interfaces render, manage state, communicate delays, handle failure, protect users, and support accessibility. Teams that plan for these behaviors early can create experiences that remain understandable and responsive even when the model output is long, delayed, incomplete, or unexpected. Flexible layouts, safe rendering, per-message state, clear feedback, and realistic performance testing form a practical foundation for dependable AI products.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1d977f860771" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How I Built and Published a Jira App on the Atlassian Marketplace]]></title>
            <link>https://concertidc.medium.com/how-i-built-and-published-a-jira-app-on-the-atlassian-marketplace-d1dae100a30b?source=rss-86328d0c504f------2</link>
            <guid isPermaLink="false">https://medium.com/p/d1dae100a30b</guid>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[app-development]]></category>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[atlassian]]></category>
            <category><![CDATA[jira]]></category>
            <dc:creator><![CDATA[ConcertIDC Engineering]]></dc:creator>
            <pubDate>Thu, 13 Aug 2026 16:49:11 GMT</pubDate>
            <atom:updated>2026-08-13T16:49:11.661Z</atom:updated>
            <content:encoded><![CDATA[<h4>A practical guide based on the development of the QA Summary Panel app.</h4><p><strong>1. Problem Statement</strong></p><p>QA tracking in Jira was taking too much manual effort.</p><p>For every Epic or parent issue, the team had to open multiple child issues, scan comments, check QA status transitions, and then manually compute the final picture: what passed, what failed, what reopened, and what was still pending. This slowed down release readiness checks and created avoidable communication gaps between QA, engineering, and stakeholders.</p><p>As this repeated sprint after sprint, I started thinking about building a focused Forge app that could do this aggregation automatically inside Jira. That idea became QA Summary panel: a panel that reads the QA Status signal from related issues, summarizes the outcome instantly, and gives teams a single place to assess quality status before release.</p><p><strong>2. The Idea &amp; Objective</strong></p><p>That’s when the idea hit me: <strong>What if Jira could show QA test results aggregated and summarized right on the issue panel?</strong></p><p>QA Summary Panel was born from a simple objective:</p><ul><li>Create a plugin that automatically aggregates QA status across all linked and child issues</li><li>Display pass/fail/reopen counts in a visual, at-a-glance format</li><li>Show team members when all tests pass with a celebratory banner</li><li>Provide real-time insights without requiring teams to change their existing Jira workflow</li></ul><p>The app’s core philosophy: <strong>Make QA visibility effortless.</strong></p><p><strong>Key Features:</strong></p><ul><li>Aggregates QA test results (PASS, FAILED, REOPEN) from the QA Status (Custom field) which will be created during the app installation.</li><li>Calculates pass rate percentage (e.g., “Pass rate: 75% · 6 of 8 reviewed”)</li><li>Shows last updated timestamp (“Updated 2 hours ago”)</li><li>Displays an “All Clear” green banner when every linked issue passes QA</li><li>Color-codes Jira statuses (Done = green, In Progress = blue, Blocked = red, To Do = grey)</li><li>Includes a refresh button for real-time updates without page reload</li><li>Works across Epic, Story, Task, and Subtask issue types</li><li>Provides customizable admin settings for team preferences</li></ul><p>The architecture is deliberately simple: <strong>the app reads the QA Status custom field, and performs aggregation logic without storing any persistent data.</strong> This design choice became crucial later when discussing security with Atlassian reviewers.</p><p><strong>3. Expected Outcomes &amp; Target Users</strong></p><p>I envisioned QA Summary panel solving problems for several personas:</p><p><strong>QA Managers:</strong></p><ul><li>Instant visibility into testing progress across sprints and epics</li><li>Ability to identify bottlenecks (issues stuck in “FAILED” state)</li><li>Quick status reports for stakeholders without manual compilation</li></ul><p><strong>Development Team Leads:</strong></p><ul><li>Real-time feedback on code quality before release</li><li>Clear signal when features are ready to move to production</li><li>Reduction in back-and-forth communication with QA team</li></ul><p><strong>Release Managers:</strong></p><ul><li>Confidence that all components in a release have been properly tested</li><li>Automated status aggregation replaces manual tracking spreadsheets</li><li>Audit trail of test transitions and timestamps</li></ul><p><strong>Jira Administrators:</strong></p><ul><li>Easy installation via Atlassian Marketplace</li><li>Minimal configuration required (just ensure QA Status field exists)</li><li>No external dependencies or infrastructure to manage</li></ul><p><strong>Expected Business Value:</strong></p><ul><li>Reduced time spent on status gathering (estimated 5–10 hours per sprint)</li><li>Fewer miscommunications about QA readiness</li><li>Faster release cycles with clear quality gates</li><li>Improved team morale with visible QA “all-clear” achievements</li></ul><p>You should design your app with <strong>clear personas in mind</strong> before building. This helps you make architectural decisions, prioritize features, and craft marketplace messaging that resonates.</p><p><strong>4. Planning and Setup: How I Help You Build and Publish YourAtlassianApp</strong></p><p>When I mentor teams building a Forge app for the first time, I keep the first phase simple: define scope, define trust boundaries, define release criteria.</p><p>For YourAtlassianApp, I recommend this planning checklist before writing implementation code:</p><ol><li>Define one measurable problem your app solves inside Jira.</li><li>Decide where the app appears: Issue Panel, Admin Page, or both.</li><li>Minimize required scopes from day one.</li><li>Decide if you need storage. If you do not need persistence, avoid it.</li><li>Define your first release as an MVP with 3 to 5 core capabilities.</li><li>Define Marketplace readiness criteria early:</li></ol><ul><li>Privacy policy, terms, logo, support channel, product documentation.</li></ul><p>My delivery style is practical: we build what can be reviewed, secured, and published quickly, then iterate in versions.</p><p><strong>Core Architecture Overview</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*KDU8QOFALIgw98zrLHI8uQ.png" /></figure><p>This structure keeps your app maintainable and review-friendly.</p><p><strong>5. Environment Setup for Beginners</strong></p><p>I guide teams to set up environment in the exact order that prevents common onboarding issues.</p><p><strong>Step 1: Install runtime and Forge CLI</strong></p><ol><li>npm install -g @forge/cli</li><li>forge — version</li></ol><p>This installs the command-line tool you will use to deploy, test, and manage your app lifecycle.</p><p><strong>Step 2: Authenticate and verify access</strong></p><ol><li>forge login — Confirm the Atlassian account you will use</li><li>Confirm the Jira site where you will test installs</li><li>Confirm you have permission to install apps on that site</li></ol><p><strong>Step 3: Prepare your local workspace</strong></p><p>Use a clean project folder and initialize dependencies:</p><ol><li>npm install</li></ol><p><strong>Step 4: Validate your release path early</strong></p><p>Before deep implementation work, verify you can execute:</p><ol><li>forge deploy</li><li>forge install</li><li>forge tunnel</li></ol><p>If this loop works, you are ready for real development.</p><p><strong>6. Clone and Scaffold YourAtlassianApp</strong></p><p>For beginners, I teach two safe paths.</p><p><strong>Path A: Start from a template scaffold</strong></p><p>Use Forge templates to bootstrap modules and manifest structure. The Forge CLI provides starter templates for common app types like issue panels and admin pages.</p><p><strong>Path B: Start from an existing repo</strong></p><p>Clone a known working baseline and evolve features module by module. This approach is faster if you have a reference app.</p><p>For YourAtlassianApp, your scaffolding should clearly separate:</p><ol><li><strong>Manifest and module declarations</strong> — manifest.yml</li><li><strong>Frontend panel rendering</strong> — React components in src/</li><li><strong>Resolver business logic</strong> — Backend handlers</li><li><strong>Admin configuration surface</strong> — Admin page for settings</li></ol><p><strong>Recommended Folder Structure</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9oM1aqa7AYUFp5MR4JOb0w.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dV9e3Uc5pDNsOyzu0Bq8nw.png" /></figure><p><strong>7. Understanding the Core Building Blocks</strong></p><p>I always explain these components first so teams avoid accidental rework.</p><p><strong>The Forge Manifest — </strong>The manifest is your contract with Atlassian:</p><ul><li><strong>Modules</strong> define app surfaces (where your app appears in Jira)</li><li><strong>Functions</strong> define backend handlers (what code runs on your behalf)</li><li><strong>Scopes</strong> define permissions (what data your app can access)</li><li><strong>Resources</strong> define UI entry points (where frontend code lives)</li></ul><p>A minimal manifest looks like this in concept:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*w0ZPDeRu7YOZC4ZigMr4IQ.png" /></figure><p><strong>The Frontend Layer</strong></p><p>Use Forge UI Kit patterns to keep UX consistent with Atlassian products. Your first version should prioritize clarity over heavy customization. Forge UI Kit 2 gives you pre-built components that match the Jira aesthetic.</p><p><strong>The Resolver Layer — </strong>This is where YourAtlassianApp should:</p><ol><li>Fetch issue context</li><li>Query related issues</li><li>Read QA Status Field signal fields</li><li>Aggregate counts and status</li><li>Return structured payload to UI</li></ol><p><strong>Security Posture from Day One</strong></p><p>Treat every scope request as something you must justify during review. Smaller scope footprint usually leads to smoother approvals.</p><p><strong>8. Coding Best Practices: How I Build Production-Ready Forge Apps for YourAtlassianApp</strong></p><p>When I guide teams building a Forge app, I focus on practices that reduce Marketplace review friction and future maintenance cost.</p><p><strong>Keep Module Boundaries Clean</strong></p><ol><li>Keep UI concerns in panel components.</li><li>Keep Jira API calls in resolver functions.</li><li>Keep transformation and aggregation logic in dedicated helper units.</li></ol><p>This makes your app easier to test and easier to audit.</p><p><strong>Design for Least Privilege</strong></p><ol><li>Start with minimum scopes.</li><li>Add new scopes only when a feature truly requires them.</li><li>Document why each scope exists.</li></ol><p>You should assume every scope might be questioned during security and review.</p><p><strong>Build for Predictable Behavior</strong></p><ol><li>Standardize response shapes between resolver and UI.</li><li>Return explicit states: success, empty, partial, error.</li><li>Handle missing fields and permissions gracefully.</li></ol><p><strong>Log Intentionally</strong></p><ol><li>Use structured logs around API calls and aggregation outcomes.</li><li>Avoid noisy logs that hide real failures.</li><li>Never log sensitive content.</li></ol><p><strong>Fail Safely in UI</strong></p><p>If resolver data is unavailable, the panel should still render a useful state:</p><ul><li>“No QA signal found yet”</li><li>“Unable to read related issues due to permissions”</li><li>“Try refresh”</li></ul><p><strong>Documentation as Part of Coding</strong></p><ol><li>Write short operational notes while implementing.</li><li>Capture assumptions and field dependencies.</li><li>Keep release notes for each deployment.</li></ol><p>You should treat docs as part of shipping, not post-shipping cleanup.</p><p><strong>9. Frontend Approach: How I Structured QA Summary panel and How You Can Build YourAtlassianApp</strong></p><p>In QA Summary panel, the frontend goal was simple: show QA readiness quickly without forcing users to navigate away from the issue.</p><p>For YourAtlassianApp, I recommend the same UI strategy.</p><p><strong>UI Principles That Worked</strong></p><ol><li>Make pass/fail/reopen counts visible first.</li><li>Show pass rate with context, not as a standalone number.</li><li>Include “last updated” to increase trust in the data.</li><li>Use status color mapping consistently.</li><li>Keep refresh explicit and fast.</li></ol><p><strong>Suggested Panel Layout</strong></p><p>The panel flows from summary to detail:</p><ol><li><strong>Summary strip:</strong> Pass, Failed, Reopen, Pending, Pass rate</li><li><strong>Signal banner:</strong> All clear vs warning state</li><li><strong>Detail section:</strong> Related issue status snapshot</li><li><strong>Action:</strong> Refresh data button</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4lDq12FFi0EQaFT0FWJCKQ.png" /></figure><p><strong>Example UI Content Snippets</strong></p><p>These are display concepts you can adapt:</p><ul><li>“Pass rate: 75% · 6 of 8 reviewed”</li><li>“Updated 2h ago”</li><li>“All clear: latest QA state is PASS across related items”</li><li>“Pending review: 2 issues have no QA result yet”</li></ul><p><strong>UX Decisions You Should Make Early</strong></p><ol><li>Decide if unresolved data should appear as pending or unknown.</li><li>Decide whether stale timestamps should show warning hints.</li><li>Decide if issue lists should be collapsed by default for large epics.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XRRLLqfzVMLaZmqoDiXuOw.png" /></figure><p><strong>10. Backend Resolver Strategy: QA Summary panel Learnings Applied to YourAtlassianApp</strong></p><p>For QA Summary panel, the resolver was the core engine. It pulled issue context, inspected linked work items, interpreted QA signals, and returned a normalized summary payload to UI.</p><p>For YourAtlassianApp, this is the repeatable backend flow.</p><p><strong>Resolver Responsibilities</strong></p><ol><li>Accept current issue context.</li><li>Identify related scope:</li></ol><ul><li>Epic children</li><li>Linked issues</li><li>Subtasks</li></ul><p>3. Read QA signal source:</p><ul><li>QA Status field</li></ul><p>4. Aggregate totals and trend markers.</p><p>5. Return compact UI-ready output.</p><p><strong>Aggregation Rules You Should Formalize</strong></p><ol><li>Use latest QA signal only per related issue.</li><li>Normalize values: PASS, FAILED, REOPEN, PENDING.</li><li>Define deterministic tie-breaking for ambiguous comment text.</li><li>Count only valid, parseable results.</li><li>Expose non-parseable cases as diagnostics.</li></ol><p><strong>Error and Edge-Case Handling</strong></p><ol><li>Missing field configuration</li><li>Linked issue permission gaps</li><li>Partial API failures</li><li>Empty relation sets</li><li>Throttling or transient API errors</li></ol><p>You should make these behaviors explicit in your design doc before coding.</p><p><strong>Example Resolver Output Contract</strong></p><p>An example response schema you can describe in documentation:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aSxTt4P2jxu-RQmwBMnGtQ.png" /></figure><p><strong>11. Data Handling Model: Why QA Summary panel Used No Storage and How to Decide for YourAtlassianApp</strong></p><p>One important design decision in QA Summary panel: no persistent storage.</p><p>The app reads QA Status signals from Jira context, computes counts at request time, and renders the result. This reduced governance overhead and simplified security conversations.</p><p>For YourAtlassianApp, use this decision rule:</p><p><strong>Use no storage when:</strong></p><ul><li>All required data already exists in Jira</li><li>Real-time computation is acceptable</li><li>You do not need historical snapshots</li></ul><p><strong>Use storage only when:</strong></p><ul><li>You need app-specific configuration</li><li>You need derived historical analytics</li><li>You need caching for performance at scale</li></ul><p><strong>Data Flow Diagram (No-Storage Model)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/154/1*yo0uSOTqKnRjzzvlAPdvUA.png" /></figure><p><strong>Security Implications of No-Storage</strong></p><ol><li>Reduced data-at-rest risk for app-managed data.</li><li>Smaller disclosure surface during review.</li><li>Easier explanation of data lifecycle: read, compute, present, discard.</li></ol><p>You should still document what is read, why it is read, and which scopes enable it.</p><p><strong>12. Testing Strategy: How I Validated QA Summary panel and How You Should Validate YourAtlassianApp</strong></p><p>Testing was a critical part of making QA Summary panel reliable before Marketplace submission. Since the app computes live summaries from Jira data, my priority was behavioral correctness across real issue structures.</p><p>For YourAtlassianApp, I recommend a testing strategy that starts manual and grows systematically.</p><p><strong>Why Manual QA Was Essential for This App</strong></p><p>In QA Summary panel, behavior depended on live Jira relationships:</p><ul><li>Epic to Story hierarchies</li><li>Subtask structures</li><li>Linked issue directions</li></ul><p>These are hard to simulate perfectly in isolation, so manual QA gave high confidence early.</p><p><strong>Manual QA Checklist You Should Adopt</strong></p><ol><li>Validate on Epic with multiple child stories.</li><li>Validate on Story with subtasks and linked issues.</li><li>Validate mixed QA outcomes: PASS, FAILED, REOPEN, missing.</li><li>Validate pass-rate math against manual calculation.</li><li>Validate “all clear” banner only when latest related QA status is PASS.</li><li>Validate timestamp freshness and refresh behavior.</li><li>Validate behavior when QA Status field is missing.</li><li>Validate behavior under permission-limited user roles.</li></ol><p><strong>Example Test Scenarios</strong></p><p><strong>Scenario 1:</strong> 8 related issues, 6 PASS, 1 FAILED, 1 no signal <br>Expected: Pass rate 75%, pending count 1, no all-clear banner.</p><p><strong>Scenario 2:</strong> All related issues latest state PASS <br>Expected: all-clear banner visible, warning state hidden.</p><p><strong>Scenario 3:</strong> Resolver cannot read one linked issue due to permissions <br>Expected: partial data flag, clear user-safe message.</p><p><strong>What to Add as Your App Matures</strong></p><p>You should start with manual QA for correctness, then add:</p><ol><li>Contract tests for resolver response shape.</li><li>Negative tests for null/missing issue fields.</li></ol><p><strong>13. Local Development Workflow: The Fast Loop I Used and You Should Follow</strong></p><p>For QA Summary panel, the fastest productive cycle was always:</p><ol><li>Update logic or UI</li><li>Deploy</li><li>Install/verify</li><li>Tunnel and iterate</li><li>Re-validate behavior in Jira</li></ol><p>For YourAtlassianApp, keep this loop strict and repeatable.</p><p><strong>Practical Development Sequence</strong></p><ol><li>Install dependencies.</li><li>Authenticate CLI.</li><li>Deploy to development environment.</li><li>Install app on test Jira site.</li><li>Run tunnel for iterative checks.</li><li>Validate on real issue combinations.</li></ol><p><strong>Example Command Snippets for Your Runbook</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wkq7QSzwR6MJE7dp_NOxOA.png" /></figure><p>After running forge tunnel, the CLI watches your local files for changes. When you modify frontend or resolver code, it automatically redeploys and you can immediately see changes in your test Jira instance without manual redeploy commands.</p><p><strong>Common Local-Development Pitfalls You Should Avoid</strong></p><ol><li>Testing only one issue type and assuming all are covered.</li><li>Skipping revalidation after scope or manifest changes.</li><li>Not testing with restricted permissions.</li><li>Treating stale issue state as fresh result without refresh.</li></ol><p><strong>Recommended Environment Discipline</strong></p><ol><li>Keep a dedicated test project in Jira.</li><li>Keep known fixture issues with predefined QA states.</li><li>Track expected versus observed values in a lightweight QA sheet.</li></ol><p><strong>14. Deploying YourAtlassianApp: Development, Validation, and Promotion</strong></p><p>In QA Summary panel, deployment was intentionally incremental to reduce release risk.</p><p>For YourAtlassianApp, use staged confidence gates.</p><p><strong>Development Deploy Gate</strong></p><p>Before every deploy:</p><ol><li>Confirm manifest changes are intentional.</li><li>Confirm scope additions are required.</li><li>Confirm visible UI labels and icon references are correct.</li></ol><p>Deploy and install only after this check.</p><p><strong>Validation Gate</strong></p><p>After deployment:</p><ol><li>Verify module placement in Jira UI.</li><li>Verify app icon and metadata render correctly.</li><li>Verify summary numbers using controlled test issues.</li><li>Verify failure messaging and refresh behavior.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1020/1*OgPJPJzqLygS1ks-jTWfwA.png" /></figure><p><strong>Promotion Readiness Checklist</strong></p><ol><li>No critical incorrect-count defects.</li><li>No scope overreach.</li><li>Marketplace assets prepared.</li><li>Legal and support pages available.</li><li>Manual QA evidence captured.</li></ol><p>You should not start Marketplace submission before these are complete.</p><p><strong>15. CI/CD and Release Hygiene: How I Kept QA Summary panel Deployable</strong></p><p>Even for a focused app, release hygiene matters. For QA Summary panel, I kept deployment operations simple but predictable.</p><p>For YourAtlassianApp, this is the baseline pipeline model.</p><p><strong>Branch and Trigger Model</strong></p><ol><li><strong>Pull request:</strong> Run dependency install and build validation.</li><li><strong>Merge to main:</strong> Deploy to development.</li><li><strong>Version tag:</strong> Deploy production-ready package.</li></ol><p><strong>Secrets and Operational Controls</strong></p><ol><li>Store Forge credentials in CI secrets.</li><li>Rotate tokens periodically.</li><li>Limit who can publish release tags.</li><li>Keep deployment logs available for auditability.</li></ol><p><strong>Release Quality Checklist</strong></p><ol><li>Changelog updated.</li><li>Reviewer-facing notes prepared.</li><li>Marketplace description aligned with actual behavior.</li><li>Icon, screenshots, and docs reflect current build.</li></ol><p><strong>Deployment Pipeline Diagram (Technical View)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/154/1*Dxbyl7Uv3dEEuAwv7_Gbvg.png" /></figure><p><strong>Why This Workflow Works in Practice</strong></p><p>This flow gives enough control to catch regressions early while staying lightweight for a small Forge team.</p><p>You should optimize for clarity and repeatability before trying to optimize for automation complexity.</p><p><strong>16. Marketplace Preparation: How I Prepared QA Summary panel and How You Should Prepare YourAtlassianApp</strong></p><p>Before submission, I treated Marketplace preparation as an engineering task, not a marketing afterthought. That made review smoother and reduced back-and-forth.</p><p>For YourAtlassianApp, I recommend preparing these in parallel with final QA.</p><p><strong>Listing-Ready Metadata</strong></p><ol><li>App name and short value statement.</li><li>Clear long description focused on user outcomes.</li><li>Supported Atlassian product scope.</li><li>Category and tags that match actual functionality.</li><li>Support contact and response expectations.</li></ol><p><strong>Visual Assets You Should Prepare</strong></p><ol><li><strong>App icon</strong> with correct dimensions and readable contrast.</li><li><strong>Product screenshots</strong> from real usage flow.</li><li><strong>Optional lightweight architecture visual</strong> for technical buyers.</li><li><strong>Consistent naming</strong> across icon, listing title, and in-product title.</li></ol><p>In QA Summary panel, one code-level feedback item during review was icon handling. The app needed to use the correct app icon expectations rather than a Jira icon style. Small branding mismatches can delay approval, so validate this early.</p><p><strong>Listing Content Quality Checks</strong></p><ol><li>Avoid vague claims.</li><li>Describe exactly what data is read and how results are shown.</li><li>Match listing screenshots to current app behavior.</li><li>Include setup prerequisites in plain language for first-time admins.</li></ol><p>You should assume reviewers will compare listing claims with observed runtime behavior.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cvk02deq2b5XUujHWqjACQ.png" /></figure><p><strong>17. Required Documentation and Assets Atlassian Expects</strong></p><p>For QA Summary panel, documentation completeness mattered almost as much as functionality. For YourAtlassianApp, this is the minimum documentation package I recommend preparing before submission.</p><p><strong>Legal and Policy Documents</strong></p><ol><li><strong>Privacy Policy:</strong> Explain what data is accessed and how it is handled.</li><li><strong>Terms and Conditions:</strong> Set user expectations and liability boundaries.</li><li><strong>Data Handling Statement:</strong> Document what is read, what is stored, and what is not stored.</li><li><strong>Security Contact Channel:</strong> Provide a way for users to report security issues.</li></ol><p><strong>Product Documentation</strong></p><ol><li><strong>Installation Guide:</strong> Step-by-step for new users.</li><li><strong>Setup Guide:</strong> How to configure required Jira fields and permissions.</li><li><strong>User Guide:</strong> Daily usage patterns and feature explanations.</li><li><strong>Troubleshooting Guide:</strong> Common issues and resolution steps.</li><li><strong>Uninstall Guide:</strong> How to safely remove the app.</li></ol><p><strong>Support and Operational Docs</strong></p><ol><li><strong>Support Channel:</strong> How users can raise issues (email, GitHub issues, support portal).</li><li><strong>Expected Response Window:</strong> Set SLA expectations.</li><li><strong>Versioning and Release Notes Model:</strong> How you communicate updates.</li><li><strong>Known Limitations Section:</strong> Be transparent about what the app does not do.</li></ol><p><strong>Documentation Checklist for YourAtlassianApp</strong></p><ol><li>Does each guide reflect the latest deployed version?</li><li>Are screenshots current?</li><li>Are field names and labels exact?</li><li>Are support and legal links live and public?</li></ol><p>You should keep all URLs stable before submission, since broken policy links can block listing progress.</p><p><strong>18. Security Review and Data Management: What I Learned from QA Summary panel</strong></p><p>Security review is where many first-time submissions slow down. I approached QA Summary panel by clearly explaining data lifecycle and permissions in reviewer-friendly terms.</p><p>For YourAtlassianApp, the strongest posture is clarity and minimalism.</p><p><strong>Data Management Model Used in QA Summary panel</strong></p><ol><li>The app reads QA signals from Jira context.</li><li>It computes counts in runtime.</li><li>It renders results to panel UI.</li><li>It does not persist business data for aggregation history.</li></ol><p>This “read, compute, present” model made security conversations more straightforward.</p><p><strong>Scope Justification Model You Should Use</strong></p><p>For each requested scope, document:</p><ol><li>Why it is needed.</li><li>Which feature depends on it.</li><li>What breaks if scope is removed.</li></ol><p>This helps you answer reviewer questions quickly and consistently.</p><p><strong>Security Questions from Atlassian Review</strong></p><p>These are the actual security and compliance questions Atlassian asked during marketplace review:</p><p><strong>Data Storage and Management</strong></p><ul><li>End-User Data processed and/or stored outside of Atlassian apps and services (excluding process/storage in logs)</li><li>Answer: The Marketplace app does not process or store End-User Data outside of Atlassian apps and services.</li></ul><p><strong>Data Residency</strong></p><ul><li>Data residency for in-scope End-User Data</li><li>Answer: Not applicable. App does not store End-User Data.</li></ul><p><strong>App REST APIs</strong></p><ul><li>Does the app provide REST APIs?</li><li>Answer: This app does not provide REST APIs.</li></ul><p><strong>In-Scope and Out-of-Scope End-User Data</strong></p><ul><li>List of in-scope End-User Data that is stored</li><li>List of out-of-scope End-User Data</li><li>Answer: Not applicable. App does not store End-User Data.</li></ul><p><strong>Integration Permissions</strong></p><ul><li>Jira administration actions (e.g., create projects and custom fields, view workflows, manage issue link types)</li><li>Create and edit project settings and create new project-level objects (e.g., versions and components)</li><li>Read field configurations</li><li>Answer: YourAtlassianApp requests permissions to read issue context, fetch related issues, and read field configurations. It does not create, edit, or delete projects, custom fields, workflows, or project settings. All permissions are read-only for data aggregation.</li></ul><p><strong>Privacy Policy</strong></p><ul><li>Is Atlassian’s privacy policy applicable to this app?</li><li>Does the app have its own privacy policy?</li><li>Answer: Atlassian’s privacy policy is not applicable to the use of this app. Please refer to the privacy policy provided by the app partner (Concert IDC).</li></ul><p><strong>GDPR Compliance</strong></p><ul><li>Is the Company/Organization a ‘data controller’ under GDPR?</li><li>Is the Company/Organization a ‘data processor’ under GDPR?</li><li>Answer: The app does not store or process personal data in a manner that would make Concert IDC a data controller or processor. Data read from Jira is computed in-memory and discarded after rendering results. No data is retained outside of Atlassian infrastructure.</li></ul><p><strong>Suggested Response Framework You Can Reuse</strong></p><ol><li>State the exact behavior.</li><li>Map behavior to scope.</li><li>Explain data flow and storage decision.</li><li>Provide user impact and security controls.</li></ol><p>You should avoid generic answers. Reviewer confidence increases when answers are concrete and feature-linked.</p><p><strong>19. Submission Process and Review Follow-Up</strong></p><p>For QA Summary panel, submission-to-publication took about two weeks, largely depending on Atlassian review queue load. The review required one code-level adjustment related to icon handling, then moved forward.</p><p>For YourAtlassianApp, this is the process I recommend documenting and following.</p><p><strong>Submission Workflow</strong></p><ol><li>Finalize listing metadata and assets.</li><li>Confirm legal links are public and valid.</li><li>Submit app for review through Marketplace partner flow.</li><li>Monitor review status and email notifications.</li><li>Respond quickly to clarification requests.</li><li>Apply fixes, redeploy, and update review notes.</li></ol><p><strong>Follow-Up Discipline That Helps</strong></p><ol><li>Keep a single tracker for all reviewer questions.</li><li>Answer each question with evidence and exact app behavior.</li><li>Include version/build references in each response.</li><li>Close the loop after fixes with concise change notes.</li></ol><p><strong>Ticketing and Status Management Pattern</strong></p><ol><li>Raise an internal tracking ticket at submission.</li><li>Log each reviewer interaction as a timeline event.</li><li>Add owner and SLA for each required action.</li><li>Close only after publication confirmation.</li></ol><p><strong>Practical Timeline Expectation</strong></p><ol><li>Initial queue and triage can vary by Atlassian load.</li><li>Plan for around two weeks as a realistic baseline.</li><li>Build buffer for at least one clarification cycle.</li><li>Treat first submission as an iterative review, not a one-shot event.</li></ol><p>You should communicate this timeline early to stakeholders so they plan release dates with approval uncertainty in mind.</p><p><strong>20. Post-Submission Operations: What I Did for QA Summary panel and What You Should Do for YourAtlassianApp</strong></p><p>Once QA Summary panel was submitted, the work did not stop. Post-submission discipline is what keeps release momentum and reviewer communication smooth.</p><p>For YourAtlassianApp, I recommend this operating model after submission:</p><ol><li>Track reviewer communications daily.</li><li>Keep one owner for reviewer response coordination.</li><li>Maintain a response log with date, question, answer, and action taken.</li><li>Keep deployment notes ready for each patch update.</li><li>Re-validate app behavior after every reviewer-driven change.</li></ol><p><strong>My Practical Post-Submission Loop</strong></p><p>For QA Summary panel, the loop was:</p><ol><li>Receive review feedback.</li><li>Assess if it is code, listing, or policy related.</li><li>Apply targeted change.</li><li>Redeploy and validate.</li><li>Reply with exact change summary.</li></ol><p>This approach kept the process predictable and reduced ambiguity.</p><p><strong>21. Versioning and Update Strategy</strong></p><p>For QA Summary panel, I treated every review-related update as a formal release event, even if the change felt small, like icon handling.</p><p>For YourAtlassianApp, follow a disciplined versioning approach from day one.</p><p><strong>Recommended Release Levels</strong></p><ol><li><strong>Patch release:</strong> Small fixes, wording updates, icon or metadata corrections.</li><li><strong>Minor release:</strong> New feature additions without breaking setup assumptions.</li><li><strong>Major release:</strong> Significant behavior changes or workflow redesign.</li></ol><p><strong>What You Should Include Per Release</strong></p><ol><li>Version number.</li><li>Change summary.</li><li>Reviewer-facing impact notes.</li><li>Validation evidence from manual QA.</li><li>Any policy/doc updates linked to that release.</li></ol><p><strong>Why This Matters</strong></p><p>During review cycles, clear version history helps Atlassian understand exactly what changed and why.</p><p><strong>22. Key Challenges I Faced Building QA Summary panel</strong></p><p>This section is intentionally candid because it can save teams time when building YourAtlassianApp.</p><p><strong>Challenge: Marketplace Review Turnaround Uncertainty</strong></p><p>The review timeline for QA Summary panel was around two weeks, influenced by Atlassian queue load. This variability can affect launch plans if not communicated early.</p><p><strong>Lesson:</strong> Build your internal timeline with a two-week baseline plus one-week buffer for clarifications. Do not promise customers a fixed launch date until the app is published.</p><p><strong>Challenge: Code-Level Feedback During Review</strong></p><p>I received one code-level request to update icon behavior so the app used the correct app icon expectation. It was a focused change, but it demonstrated how detail-oriented review can be.</p><p><strong>Lesson:</strong> Validate icon formats, manifest references, and UI consistency against official Atlassian guidelines before submission. Small details like this can delay publication.</p><p><strong>Challenge: Enterprise Onboarding Conversations</strong></p><p>As part of onboarding and publication readiness, I had to address enterprise-facing concerns around trust, security clarity, and operational maturity.</p><p><strong>Lesson:</strong> Prepare detailed security and operational documentation early. Enterprise buyers want proof of maturity, clear data handling, and responsive support.</p><p><strong>Challenge: Security Question Handling</strong></p><p>I prepared clear technical explanations for data access and scope use. Security reviewers want concrete answers tied to actual code behavior.</p><p><strong>Lesson:</strong> Review your scopes ruthlessly. Every permission should have a clear feature mapping. If you cannot justify a scope, remove it. Minimal scope footprint leads to faster approval.</p><p><strong>23. Lessons Learned and Recommendations</strong></p><p>Building QA Summary panel taught me that technical quality alone is not enough for Marketplace success. Submission readiness is a combination of engineering clarity, documentation quality, and response speed.</p><p>For YourAtlassianApp, here are the most useful lessons:</p><ol><li><strong>Keep your first version narrow and dependable.</strong> Scope the MVP to one clear problem, not five. Marketplace reviewers respect focused apps.</li><li><strong>Minimize permission scopes and justify each one.</strong> Every scope is a security conversation. Start with nothing and add only what is required.</li><li><strong>Prefer simple data models, especially if no storage is needed.</strong> No storage means lower governance overhead, faster security review, and fewer compliance questions.</li><li><strong>Build docs and legal assets in parallel with coding.</strong> Do not treat documentation as post-shipping cleanup. Write it as you code.</li><li><strong>Expect at least one review clarification cycle.</strong> Plan for iteration. First submission rarely goes straight to publish.</li><li><strong>Keep your reviewer responses technical and specific.</strong> Avoid generic answers. Reference code, manifests, and actual behavior. This builds confidence.</li><li><strong>Treat small UI and metadata details as release-critical.</strong> Icon format, field labels, and screenshot accuracy matter. They signal production readiness.</li></ol><p>You should plan for publication as part of development, not as a final checklist at the end.</p><p><strong>24. Quick Reference Appendix for YourAtlassianApp</strong></p><p><strong>24.1 Command Runbook</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6IOuIMQMSieBbzOdUGw--g.png" /></figure><p><strong>24.2 Build-to-Publish Checklist</strong></p><p>Development:</p><p>​​☐​ App functions validated in Jira test project</p><p>​​☐​ Manual QA checklist completed</p><p>​​☐​ No console errors or warnings in tunnel</p><p>Pre-Submission:</p><p>​​☐​ Icon and screenshots aligned with current app</p><p>​​☐​ Privacy policy and terms published and accessible</p><p>​​☐​ User guide and setup guide published</p><p>​​☐​ Support contact configured and monitored</p><p>​​☐​ Submission package reviewed by internal stakeholders</p><p>Submission:</p><p>​​☐​ Listing metadata finalized</p><p>​​☐​ App ID and version recorded</p><p>​​☐​ Reviewer contact assigned</p><p><strong>24.3 Reviewer Response Template</strong></p><p>When you receive reviewer feedback, use this structure:</p><ol><li><strong>Question summary:</strong> Restate what was asked.</li><li><strong>Exact technical behavior:</strong> Describe what the code does.</li><li><strong>Scope/data mapping:</strong> Link behavior to requested permissions.</li><li><strong>Fix details (if any):</strong> Explain what changed and why.</li><li><strong>Version and deployment reference:</strong> Include build number and re-deployment timestamp.</li></ol><p>Example:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ERMB9KT1MUrAeUr4Js3z4A.png" /></figure><p><strong>25. References and Further Reading</strong></p><p>For a technical audience building YourAtlassianApp, these Forge documentation areas are the most relevant:</p><p><strong>Official Atlassian Documentation</strong></p><ol><li><a href="https://developer.atlassian.com/platform/forge/getting-started/">Forge Getting Started</a></li><li><a href="https://developer.atlassian.com/platform/forge/build-a-hello-world-app-in-jira/">Jira App Module Guides</a></li><li><a href="https://developer.atlassian.com/platform/forge/ui-kit/">Forge UI Kit Documentation</a></li><li><a href="https://developer.atlassian.com/platform/forge/permissions/">Forge Permissions and Scopes</a></li><li><a href="https://developer.atlassian.com/platform/forge/shared-responsibility-model/">Forge Shared Responsibility Model</a></li><li><a href="https://developer.atlassian.com/platform/forge/environments-and-deployment/">Forge Deployment and Environments</a></li><li><a href="https://marketplace.atlassian.com/">Atlassian Marketplace Listing Guidelines</a></li></ol><p><strong>Jira REST API Reference</strong></p><ul><li><a href="https://developer.atlassian.com/cloud/jira/platform/rest/v3/">Jira Cloud REST API</a></li></ul><p><strong>End-to-End Process Map</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/154/1*rtrBOZ4DHTbAFT6gT8xoww.png" /></figure><p><strong>Closing Guidance</strong></p><p>I built QA Summary panel to solve a practical QA visibility problem inside Jira. If you follow this same engineering and submission discipline, you can build and publish YourAtlassianApp with fewer surprises and a much smoother review experience.</p><p>The Atlassian Marketplace is open to apps that are well-engineered, clearly documented, and secure by design. Your app is ready when it demonstrates all three.</p><p><strong>About the Author</strong></p><p>QA Summary panel was built as a focused solution to improve QA visibility in Jira. The experience of building, submitting, and publishing a Forge app taught lessons in security clarity, documentation discipline, and marketplace readiness that apply to any Atlassian app.</p><ul><li><strong>License:</strong> This guide is published by ConcertIDC and is available for reference and educational purposes.</li></ul><p><strong>Support Contact:</strong> <a href="/service/mailto:apps@concertidc.com">apps@concertidc.com</a></p><p><strong>Company Website:</strong> <a href="https://www.concertidc.com/">www.concertidc.com</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d1dae100a30b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What is Unit Testing?]]></title>
            <link>https://concertidc.medium.com/what-is-unit-testing-f2683589c61a?source=rss-86328d0c504f------2</link>
            <guid isPermaLink="false">https://medium.com/p/f2683589c61a</guid>
            <category><![CDATA[unit-testing]]></category>
            <dc:creator><![CDATA[ConcertIDC Engineering]]></dc:creator>
            <pubDate>Mon, 24 Feb 2025 20:28:09 GMT</pubDate>
            <atom:updated>2025-02-24T20:28:09.074Z</atom:updated>
            <content:encoded><![CDATA[<p>Unit testing is a type of software testing where individual units or small pieces of code are tested in isolation. The goal is to ensure that each part of the code behaves as expected in a controlled environment.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*x-aDwf05RtPiRmzw" /></figure><p>Imagine you’re building a house. You would check each material for quality and ensure that every part works together to create a strong, reliable structure. Similarly, in software development, unit testing ensures that individual units of code function correctly before they are integrated into a larger system.</p><p>In essence, unit testing is about verifying that small “units” of code work correctly on their own before they interact with other parts of the application.</p><h3>Why Do We Need Unit Testing?</h3><p>Unit testing helps ensure code reliability, providing greater confidence and control over the final application when all parts are integrated.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*yJsy7SZc8-xcA3vj" /></figure><p>Here are some key benefits:</p><ol><li><strong>Catch Bugs Early : </strong>Testing each piece of code individually helps catch bugs during development. By running tests automatically with Git commit hooks, you can prevent buggy code from reaching production. This saves time by reducing debugging efforts in live environments. <em>Example:</em> If you write a function that calculates a user’s age, a unit test could ensure it correctly handles edge cases like leap years or future birthdates before merging into the main branch.</li><li><strong>Acts as Documentation : </strong>Unit tests serve as living documentation for your code. They define how each function or component is expected to behave, making it easier for new developers to understand the system without extensive explanations.</li><li><strong>Improves Code Quality : </strong>Writing unit tests encourages better software design and cleaner code. Since unit testing requires modular and testable code, developers naturally adopt best practices such as separation of concerns and single responsibility.</li><li><strong>Helps with Refactoring and Scalability : </strong>With a solid unit testing suite, developers can confidently refactor or modify existing code without the fear of introducing new bugs. This makes scaling applications much easier.</li></ol><h3>How Can You Add Unit Testing to Your React Project?</h3><p>There are various testing tools available, but <strong>Jest</strong> and <strong>React Testing Library</strong> are the most commonly used for React projects. Here’s why they are popular:</p><ul><li><strong>Jest</strong> is a powerful testing framework that includes test runners, assertion libraries, and built-in mocking capabilities. It’s optimized for speed and works well with JavaScript applications.</li><li><strong>React Testing Library (RTL)</strong> focuses on testing how components behave rather than their internal implementation. It encourages best practices by simulating real user interactions instead of relying on component internals.</li></ul><h3>Jest vs. React Testing Library</h3><ul><li><strong>Jest</strong> is a general-purpose testing framework that can be used for unit, integration, and snapshot testing.</li><li><strong>React Testing Library</strong> is specifically designed for testing React components by rendering them and checking expected outputs rather than implementation details.</li></ul><h3>Alternatives to Jest and React Testing Library:</h3><ul><li>Instead of Jest, you could use <strong>Vitest</strong> (a faster alternative, optimized for modern JavaScript/TypeScript projects).</li><li>Instead of React Testing Library, you could use <strong>Enzyme</strong> (an older React testing library) or <strong>Cypress</strong> (which is more suited for end-to-end testing).</li></ul><h3>How to Write Tests? (Best Practices)</h3><ol><li><strong>Test Behavior, Not Implementation : </strong>A good unit test focuses on <em>what</em> the component does rather than <em>how</em> it works internally. This keeps tests stable even when refactoring the implementation. <em>Example:</em> Instead of testing the internal state of a component, check if the correct UI is displayed when a user interacts with it.</li></ol><p><strong>2. Write Clear and Descriptive Test Cases: </strong>Each test case should have a meaningful name that describes what it verifies. <em>Example:</em></p><ul><li>✅ Good: should display an error message when input is empty</li><li>❌ Bad: should display something when input is empty</li></ul><p>3. <strong>Aim for Small and Isolated Tests: </strong>Each test case should focus on one specific behavior to reduce dependencies and improve readability. <em>Example: </em>Test a button click separately from a form submission to ensure both actions are tested independently.</p><p>4. <strong>Keep Tests Maintainable: </strong>Organize tests logically, use helper functions for repeated actions, and avoid redundant code.</p><h3><strong>How Not to Write Tests? (Common Pitfalls)</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ot3tu4wJ5pty0fYm" /></figure><ol><li><strong>Avoid Testing Implementation Details </strong>Don’t test internal logic, such as state updates or private methods. Instead, test the expected behavior and outcomes. <em>Example:</em></li></ol><p>❌ Testing internal state changes of a React component.</p><p>✅ Testing whether a button click updates the UI correctly.</p><p>2. <strong>Don’t Overuse Mocks: </strong>Excessive use of mocks can lead to unrealistic tests. Use real props and actual function calls whenever possible. <em>Example:</em></p><p>❌ Mocking a function unnecessarily instead of testing the real logic.</p><p>✅ Using an actual function and verifying its output.</p><p><strong>2. Don’t Write Tests for Everything: </strong>Not all code needs tests — especially simple functions with trivial behavior. Writing too many tests slows down development and increases maintenance costs</p><p><em>Example:</em> You don’t need to test a utility function like sum(a, b) since its correctness is obvious.</p><p>✅ <em>A good practice:</em> Aim for meaningful test coverage rather than blindly targeting 100%.</p><p><strong>3. Avoid Flaky (Unstable) Tests</strong></p><p>Tests that break frequently due to minor code changes are a sign of tightly coupled test logic. To make tests more resilient, focus on behavior instead of specific implementation details. <em>Example: </em>❌ A test that fails when a button’s class changes. ✅ A test that checks if clicking the button triggers the expected action.</p><h3>Conclusion</h3><p>Unit testing may seem like extra work at first, but it is a long-term investment that pays off by improving code reliability, maintainability, and developer confidence.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/709/1*tX7QKO-3JWzUp0H2C85LrQ.png" /></figure><p>In our project, we initially had no unit tests, but after adopting them, we saw a significant improvement in code quality and development speed. Now, debugging is easier, and we can refactor code with confidence.</p><p>Most companies enforce unit testing as part of their development process, with strict code coverage requirements before approving and merging code. As software development evolves, writing tests is no longer optional — it’s a necessity.</p><p>By following best practices and avoiding common pitfalls, you can make unit testing an integral part of your development workflow without slowing down progress.</p><h3>Ram Kumar Barani</h3><p>Technical Lead</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f2683589c61a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How Vector Databases Revolutionize the Handling of Unstructured Data]]></title>
            <link>https://concertidc.medium.com/how-vector-databases-revolutionize-the-handling-of-unstructured-data-7655f356b5b6?source=rss-86328d0c504f------2</link>
            <guid isPermaLink="false">https://medium.com/p/7655f356b5b6</guid>
            <category><![CDATA[genai]]></category>
            <category><![CDATA[vector-database]]></category>
            <category><![CDATA[vector]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[rdbms]]></category>
            <dc:creator><![CDATA[ConcertIDC Engineering]]></dc:creator>
            <pubDate>Mon, 24 Feb 2025 08:57:44 GMT</pubDate>
            <atom:updated>2025-02-24T08:57:44.402Z</atom:updated>
            <content:encoded><![CDATA[<p>In the era of artificial intelligence (AI) and machine learning, the amount of unstructured data — such as images, videos, text, and audio — has skyrocketed. Traditional databases, designed to handle structured data like numbers and strings, fall short when it comes to efficiently processing and querying such complex data types. Enter vector databases: a revolutionary approach to storing, indexing, and querying high-dimensional vector representations of unstructured data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JrzyuoHfugSSxvUGFvGwhA.png" /></figure><h4>Traditional Databases vs. Vector Databases</h4><p>Traditional databases, such as relational databases (e.g., MySQL, PostgreSQL), are designed to store structured data organized into tables with rows and columns. They excel at handling numerical or categorical data and support SQL-based querying. However, they struggle with:</p><p>· Storing high-dimensional data.</p><p>· Performing similarity searches.</p><p>· Handling unstructured data like images, videos, or text.</p><p>Vector databases, on the other hand, are purpose-built to manage unstructured data encoded as high-dimensional vectors. These databases specialize in similarity searches and efficiently retrieve data points based on their proximity in vector space. This makes them ideal for applications requiring semantic understanding and pattern recognition.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MBss5HX1c2x9wxj4GEmevQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_fJXJvJ5XnkA9JrDFL7EvA.png" /></figure><p><strong>The Challenge of Unstructured Data</strong></p><p>Unstructured data makes up a significant portion of the data generated today. Examples include:</p><ul><li><strong>Text</strong>: Emails, social media posts, articles, and chat logs.</li><li><strong>Images</strong>: Photos, illustrations, and scanned documents.</li><li><strong>Audio</strong>: Voice recordings, music, and podcasts.</li><li><strong>Videos</strong>: Movies, surveillance footage, and user-generated content.</li></ul><p>Unlike structured data, which fits neatly into rows and columns, unstructured data requires more sophisticated methods to interpret and analyze its content. This is where vector embeddings and vector databases come into play.</p><p><strong>What Are Vector Databases?</strong></p><p>Vector databases store data in the form of vectors — numerical representations of unstructured data. These vectors are created by AI models that process the original data, encoding its features into high-dimensional spaces. For instance:</p><ul><li>A sentence can be transformed into a vector using models like BERT or OpenAI embeddings.</li><li>An image can be represented as a feature vector extracted by a convolutional neural network (CNN).</li><li>Audio files can be encoded into vectors through spectrogram analysis.</li></ul><p>Each vector represents a data point’s position in a high-dimensional space, where the distance between vectors indicates their similarity. For example, similar images or semantically related sentences will have vectors that are close together.</p><p><strong>How Vector Databases Transform Unstructured Data Handling</strong></p><p>Vector databases are specifically designed to handle the complexities of unstructured data, offering several key advantages:</p><p><strong>1. Efficient Similarity Search</strong>:</p><ul><li>Traditional databases struggle with finding similar data points, but vector databases excel by leveraging similarity metrics like cosine similarity or Euclidean distance.</li><li>Example: Retrieving visually similar images or semantically related text snippets.</li></ul><p><strong>2. Scalability</strong>:</p><ul><li>Vector databases can scale horizontally to store and query billions of vectors, making them ideal for applications with massive datasets.</li></ul><p><strong>3. Advanced Indexing</strong>:</p><ul><li>Technologies like Hierarchical Navigable Small World (HNSW) graphs, Annoy, and K-D Trees enable fast and efficient nearest neighbor searches.</li></ul><p><strong>4. Multimodal Data Handling</strong>:</p><ul><li>Vector databases allow seamless querying across different types of data, such as finding images based on text descriptions or retrieving audio clips similar to a sample input.</li></ul><p><strong>5. Real-Time Insights</strong>:</p><ul><li>They support real-time processing and retrieval, making them invaluable for applications like recommendation systems, chatbots, and fraud detection.</li></ul><p><strong>Real-World Applications</strong></p><ol><li><strong>Semantic Search</strong>:</li></ol><ul><li>Platforms like search engines or e-commerce sites use vector databases to retrieve results based on meaning rather than exact keyword matches.</li></ul><p><strong>2. Recommendation Systems</strong>:</p><ul><li>Streaming services and online retailers analyze user preferences to suggest personalized content or products.</li></ul><p><strong>3. Anomaly Detection</strong>:</p><ul><li>Vector databases help identify unusual patterns in financial transactions, cybersecurity events, or quality control processes.</li></ul><p><strong>4. AI-Powered Tools</strong>:</p><ul><li>From chatbots to virtual assistants, vector databases enable AI models to retrieve relevant context and improve interaction quality.</li></ul><p><strong>Popular Vector Databases</strong></p><ul><li><strong>Pinecone</strong>: A fully managed service for large-scale vector search.</li><li><strong>Weaviate</strong>: Open-source and ideal for semantic search use cases.</li><li><strong>Milvus</strong>: Built for scalability and high-performance vector retrieval.</li><li><strong>Vespa</strong>: Integrates vector search with traditional search capabilities.</li></ul><p><strong>Conclusion</strong></p><p>Vector databases are transforming how we handle unstructured data by leveraging the power of AI-driven vector embeddings. As AI models become more advanced and data continues to grow in complexity, vector databases will play an increasingly vital role in applications ranging from search engines to intelligent assistants. For organizations looking to stay ahead in the AI revolution, adopting vector databases is no longer optional — it’s essential.</p><h3>Karthiyayini Muthuraj</h3><p><em>Senior Technical Lead, </em><strong>ConcertIDC</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7655f356b5b6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Meta’s Llama 3.2: A Leap Forward in Multimodal AI]]></title>
            <link>https://concertidc.medium.com/metas-llama-3-2-a-leap-forward-in-multimodal-ai-ffb48a43db65?source=rss-86328d0c504f------2</link>
            <guid isPermaLink="false">https://medium.com/p/ffb48a43db65</guid>
            <category><![CDATA[llama-3]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[multimodal-ai]]></category>
            <category><![CDATA[multimodal]]></category>
            <category><![CDATA[ai-model]]></category>
            <dc:creator><![CDATA[ConcertIDC Engineering]]></dc:creator>
            <pubDate>Mon, 16 Dec 2024 09:17:20 GMT</pubDate>
            <atom:updated>2024-12-16T09:17:20.015Z</atom:updated>
            <content:encoded><![CDATA[<p>Meta AI continues to push the boundaries of artificial intelligence with the release of Llama 3.2, a state-of-the-art multimodal model designed to handle both text and visual inputs seamlessly. Building on the success of the Llama 3 family, this release underscores Meta AI’s commitment to innovation in large language models (LLMs) and multimodal capabilities.</p><p>Here’s an in-depth look at what makes Llama3.2 a game-changer in the world of AI.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/621/1*bB9ZeZyGl5BQiAu2fAplzQ.png" /></figure><p><strong>What Is Llama3.2?</strong></p><p>Llama 3.2 is a multimodal model, meaning it can process and generate outputs based on both textual and visual data. Unlike its predecessors, which focused primarily on text-based applications, Llama 3.2 incorporates advanced image understanding capabilities, making it</p><p>Versatile across various domains such as content creation, virtual assistants, customer support, and more.</p><p><strong>Key Features of Llama3.2</strong></p><p><strong><em>1. Multimodal Input Processing</em></strong></p><p>Llama3.2 can interpret and integrate visual and textual information. For instance:</p><ul><li>It can analyze images and answer questions about them (e.g., identifying objects, describing scenes).</li><li>It can combine image data with text for more complex tasks like diagram explanation or visual storytelling.</li></ul><p><strong><em>2. Enhanced Natural Language Understanding</em></strong></p><p>Meta AI has refined the model’s linguistic capabilities, ensuring more accurate and contextually appropriate responses. Llama 3.2 exhibits significant improvements in understanding idiomatic expressions, complex queries, and nuanced contexts.</p><p><strong><em>3. Efficient Training and Fine-Tuning</em></strong></p><p>Trained on an optimized mixture of datasets, Llama 3.2 balances general knowledge with specialized understanding. Its fine-tuning capabilities allow businesses to customize the model for specific applications, whether in healthcare, education, or e-commerce.</p><p><strong><em>4. Improved Visual Comprehension</em></strong></p><p>The model excels in tasks like:</p><ul><li>Recognizing patterns and anomalies in data visualizations.</li><li>Interpreting info graphics, charts, and images with textual overlays.</li><li>Understanding complex visual layouts such as multi-page documents.</li></ul><p><strong><em>5. Robust Multilingual Support</em></strong></p><p>Llama 3.2 supports multiple languages, making it ideal for global applications. It also adapts to regional dialects and colloquialisms, ensuring inclusivity in communication.</p><p><strong><em>6. Scalability and Energy Efficiency</em></strong></p><p>Meta AI has focused on making Llama 3.2 both powerful and sustainable. Its architecture is optimized for deployment on diverse hardware setups, from cloud servers to edge devices, with reduced energy consumption.</p><p><strong>Applications of Llama3.2</strong></p><p><strong><em>Education and Training</em></strong></p><ul><li>Interactive Learning: Create dynamic lessons with text and visual aids.</li><li>Automated Grading: Analyze student submissions with text and visual elements, such as essays and diagrams.</li></ul><p><strong><em>Healthcare</em></strong></p><ul><li>Diagnostic Support: Assist in interpreting medical images and combining findings with patient histories.</li><li>Patient Interaction: Answer queries with a mix of visual and textual information.</li></ul><p><strong><em>Retail and E-Commerce</em></strong></p><ul><li>Visual Search: Help users find products by analyzing uploaded images.</li><li>Enhanced Chatbots: Combine visual recommendations with conversational AI.</li></ul><p><strong><em>Content Creation</em></strong></p><ul><li>Visual Storytelling: Generate narratives complemented by relevant visuals.</li><li>Design Assistance: Provide contextual suggestions for graphic design or video editing projects.</li></ul><p><strong>Challenges and Opportunities</strong></p><p><strong><em>Ethical Considerations</em></strong></p><p>As with any AI model, ensuring ethical use is paramount. Meta AI has implemented safeguards to prevent misuse, but developers and users must remain vigilant in addressing potential biases or inaccuracies.</p><p><strong><em>GDPR Compliance</em></strong></p><p>For enterprises in regions with strict data regulations, Llama 3.2 provides tools for anonymization and secure data handling, ensuring compliance with standards like GDPR.</p><p><strong><em>Accessibility</em></strong></p><p>By supporting multiple modalities and languages, Llama 3.2 opens up new opportunities for accessibility, making technology more inclusive for individuals with disabilities or those in under served regions.</p><p><strong>The Road Ahead</strong></p><p>Meta AI’s Llama 3.2 sets a new benchmark for multimodal models, paving the way for a future where AI interacts seamlessly across mediums. Its robust capabilities signal a significant leap in how we approach AI-driven solutions, fostering innovation across industries.</p><p>As organizations begin to explore the possibilities of Llama 3.2, it’s clear that this multimodal model is not just an upgrade but a glimpse into the transformative potential of AI.</p><p>Stay tuned as the world of AI continues to evolve, with Llama 3.2 leading the charge toward smarter, more versatile, and more ethical applications.</p><h3>Karthiyayini Muthuraj</h3><p><em>Senior Technical Lead, </em><strong>ConcertIDC</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ffb48a43db65" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Exploring Graph Theory through Tile Patterns]]></title>
            <link>https://concertidc.medium.com/exploring-graph-theory-through-tile-patterns-065163df37d2?source=rss-86328d0c504f------2</link>
            <guid isPermaLink="false">https://medium.com/p/065163df37d2</guid>
            <category><![CDATA[math]]></category>
            <category><![CDATA[graph-theory]]></category>
            <category><![CDATA[graph]]></category>
            <category><![CDATA[graph-theory-simplified]]></category>
            <dc:creator><![CDATA[ConcertIDC Engineering]]></dc:creator>
            <pubDate>Sun, 15 Dec 2024 12:05:51 GMT</pubDate>
            <atom:updated>2024-12-16T07:17:03.866Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*SJx51y7_a286WEV4" /></figure><h3>What is Graph Theory?</h3><p>Graph theory is the study of <strong>networks</strong> made up of points (called <strong>vertices</strong>) and lines connecting them (called <strong>edges</strong>). While graphs might sound abstract, they exist all around us — in road networks, computer networks, and even the tile patterns on the floor!</p><h3>Tiles as a Graph</h3><p>Take a closer look at the tile pattern in the image. Instead of viewing the tiles as shapes, let’s think about them as a <strong>graph</strong>:</p><ol><li><strong>Vertices</strong>: Each tile is represented as a <strong>vertex</strong> (a point in the graph).</li><li><strong>Edges</strong>: If two tiles share a boundary (a side), draw an edge between their vertices.</li></ol><p>This turns the pattern into a graph where the connections between tiles become edges.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*I2D2uZr33P1sIN7X_JEbzw.png" /></figure><h3>Euler’s Formula and the Tile Graph</h3><p>Leonhard Euler discovered a cool relationship between vertices (V), edges(E), and faces/regions(R) of a <strong>connected graph</strong>. It is called the <strong>Euler’s Characteristic</strong>:</p><h3>V−E+R=2</h3><p>Here:</p><ul><li>V is the number of vertices.</li><li>E is the number of edges.</li><li>R is the number of faces/regions.</li></ul><h3>How Does This Apply ?</h3><p>In the image:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7XVVe0K7awMOSjMP0UqPeg.png" /></figure><ul><li>There are <strong>8 tiles</strong>, so we have <strong>8 vertices</strong> (V=8).</li><li>Tiles that share boundaries are connected, creating <strong>11 edges</strong> (E=11).</li><li>Since this graph is drawn on a flat surface, there is <strong>1 outer region</strong> that surrounds the entire graph and there are 4 inner regions. (R=5).</li><li>Plug these numbers into the Euler’s formula to see if V−E+R=2</li></ul><h3>Try It Yourself!</h3><ol><li><strong>Step 1</strong>: Pick a part of the tile pattern. You can draw it on paper or trace it.</li><li><strong>Step 2</strong>: Count the vertices, edges, and faces/regions.</li><li><strong>Step 3</strong>: Check Euler’s formula. Does V−E+R equal 2?</li></ol><p>This fun activity will help you explore graph theory and Euler’s characteristic while observing real-world patterns!</p><h3>Why this matters ? and what is the relevance ?</h3><p>Euler’s characteristic is not just about polygons or tile patterns — it’s a <strong>fundamental concept</strong> in math and science. This simple idea of treating tiles as vertices and shared edges as connections has far-reaching applications:</p><ol><li><strong>Network Design</strong>: Like internet connections or social networks — vertices are computers/people, and edges are connections between them.</li><li><strong>City Planning</strong>: Roads (edges) connect locations (vertices).</li><li><strong>AI and Data Structures</strong>: Graphs are at the heart of building systems like recommendation engines or path finding algorithms.</li></ol><p>It’s like finding order in chaos! Euler’s characteristic shows that even complex patterns have structure and follow rules. It connects math to the real world. Euler’s rule is used everywhere, from bridges to video games.</p><h3>Conclusion</h3><p>The tile pattern in the image isn’t just a design —it’s a graph waiting to be explored. By treating tiles as vertices and connections as edges, we can uncover the hidden order and structure in what looks like a simple floor design. Math is everywhere — sometimes, you just have to look down to see it!</p><p>Whether it’s <strong>exploring connections</strong>, <strong>solving puzzles</strong>, or building AI systems, graph theory gives us tools to see patterns and relationships in ways that go far beyond the classroom. So, the next time you walk across a tiled floor, remember — you’re walking on a graph!</p><p>Now it’s your turn to explore and discover how math is hidden all around us!</p><h3>Kannan K</h3><p>(Volunteer at RAM Foundation)</p><p><em>Director of Technology, </em><strong>ConcertIDC</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=065163df37d2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Understanding Generative AI: Unlocking the Power of Large Language Models]]></title>
            <link>https://concertidc.medium.com/understanding-generative-ai-unlocking-the-power-of-large-language-models-b70391f18d20?source=rss-86328d0c504f------2</link>
            <guid isPermaLink="false">https://medium.com/p/b70391f18d20</guid>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[large-language-models]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[genai]]></category>
            <category><![CDATA[generative-ai-use-cases]]></category>
            <dc:creator><![CDATA[ConcertIDC Engineering]]></dc:creator>
            <pubDate>Fri, 15 Nov 2024 14:57:35 GMT</pubDate>
            <atom:updated>2024-11-15T14:57:35.272Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/622/1*YKyTuOi3oj9Rzsmg8YcymQ.png" /><figcaption>Fun fact: this is an AI Generated Image</figcaption></figure><p>Artificial Intelligence (AI) has transformed industries, from healthcare to entertainment, but a particular branch called Generative AI is making waves by doing something incredibly unique — creating new content. Whether it’s generating text, producing art, or composing music, Generative AI is reshaping the boundaries of machine creativity.</p><p>Let’s explore what Generative AI is, the transformative power of Large Language Models (LLMs), and how it differs from traditional AI techniques.</p><p><strong>What is Generative AI?</strong></p><p>Generative AI refers to a type of artificial intelligence focused on producing new, original data, such as text, images, audio, or even video, based on patterns it has learned from existing datasets. Unlike traditional AI systems that perform tasks like classification or regression (e.g., identifying objects in an image or predicting stock prices), Generative AI goes beyond analysis and instead synthesizes novel content.</p><p>A well-known example is <strong>ChatGPT</strong>, which can generate human-like text based on user prompts. The capabilities of generative models make them highly adaptable for various applications, from chatbots and virtual assistants to more creative domains like graphic design and music composition.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/735/1*GOxYjtlKHiT0JusYjH2VZA.png" /></figure><h3>The Power of Large Language Models (LLMs)</h3><p>At the heart of Generative AI’s remarkable success is the Large Language Model (LLM). LLMs are deep learning models that have been trained on vast amounts of text data to understand, generate, and manipulate natural language. The real strength of LLMs lies in their ability to capture complex patterns and relationships in language through millions or billions of parameters.</p><p><strong>Here’s why LLMs are so powerful:</strong></p><p>1.<strong> Contextual Understanding:</strong> LLMs are designed to comprehend context over long pieces of text. They are not limited to generating text based on immediate prompts but also recognize nuanced relationships between words, phrases, and concepts. For instance, LLMs like GPT-4 can write coherent essays, explain intricate scientific topics, or engage in philosophical debates by leveraging deep contextual awareness.</p><p>2. <strong>Transferability Across Domains: </strong>Once trained on diverse datasets, LLMs exhibit the ability to perform well across multiple domains, including those they weren’t specifically trained for. They can generate poetry, summarize legal documents, or write code — showing immense versatility in understanding and generating domain-specific content.</p><p>3. <strong>Massive Scale and Fine-Tuning: </strong>LLMs can be trained on massive datasets, which makes them extremely well-versed in various topics. Their scale allows them to learn and generate highly detailed, accurate, and contextually relevant outputs. Moreover, finetuning these models for specific tasks further enhances their accuracy and relevance to niche applications, such as legal text generation or medical diagnosis.</p><h3>How Generative AI Differs from Other AI Techniques</h3><p>While Generative AI has garnered immense attention, it’s crucial to understand how it differs from other types of AI techniques. Here are some key differences:</p><p><strong>1. Generative vs. Discriminative Models:</strong> Traditional AI models like decision trees, support vector machines, or deep learning classifiers are generally <strong>discriminative models</strong>. These models focus on distinguishing between different categories or classes. For instance, a discriminative model would predict whether an email is spam or not. <strong>Generative models,</strong> on the other hand, create new data points by understanding the underlying distribution of the data. A generative model could, for example, compose a unique email that appears similar to a human-written message.</p><p>2. <strong>Task-Specific vs. Creative:</strong> Conventional AI models are often task-specific, designed to solve specific problems like facial recognition, fraud detection, or language translation. Generative AI is not confined to a single task. It is inherently <strong>creative</strong>, producing novel outputs that did not exist in the training data. It’s not just solving problems; it’s creating new solutions.</p><p>3.<strong> Supervised vs. Unsupervised Learning: </strong>Many traditional AI methods rely on <strong>supervised learning</strong>, where the model is trained on labeled data. Generative AI models, especially those like LLMs, often use <strong>unsupervised </strong>or<strong> self-supervised learning</strong>, where they are trained on vast amounts of raw, unlabeled data. This allows them to understand broader patterns in data without explicit human guidance, enabling them to generate coherent, human-like content.</p><p>4.<strong> Static vs. Dynamic Interaction: </strong>Traditional AI models typically perform static tasks — once a decision is made, the interaction ends. Generative AI, especially models like conversational agents, are <strong>dynamic</strong>. They engage in multi-turn dialogues, allowing users to ask follow-up questions, clarify points, and refine their prompts for a more nuanced response.</p><h3>Conclusion</h3><p>Generative AI represents a paradigm shift in how machines interact with data and produce new content. With the power of <strong>Large Language Models</strong> and their capacity to handle context, scale, and complexity, the possibilities for innovation are enormous. Whether it’s transforming how businesses communicate with customers, aiding in creative processes, or developing virtual worlds, Generative AI is just beginning to unlock its full potential. Understanding its distinctions from other AI techniques only highlights its role as a trailblazer in the next generation of machine intelligence.</p><h3>Karthiyayini Muthuraj</h3><p><em>Senior Technical Lead, </em><strong>ConcertIDC</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b70391f18d20" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Project SID: Advancing AI Collaboration and Social Structures in the Minecraft Universe]]></title>
            <link>https://concertidc.medium.com/project-sid-advancing-ai-collaboration-and-social-structures-in-the-minecraft-universe-4787cad37b5b?source=rss-86328d0c504f------2</link>
            <guid isPermaLink="false">https://medium.com/p/4787cad37b5b</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[agenticframework]]></category>
            <category><![CDATA[proejctsid]]></category>
            <category><![CDATA[ai-agents-army]]></category>
            <category><![CDATA[agents]]></category>
            <dc:creator><![CDATA[ConcertIDC Engineering]]></dc:creator>
            <pubDate>Sat, 07 Sep 2024 09:17:58 GMT</pubDate>
            <atom:updated>2024-09-07T09:25:28.008Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/749/1*y3_c1NgZEC55KXK0v5gtkQ.png" /></figure><p>Recent advancements in <strong>Project SID</strong>, set within the Minecraft universe, are pushing the boundaries of artificial intelligence (AI) and virtual collaboration. The project simulates over 1,000 autonomous AI agents within a Minecraft server, enabling them to develop complex social structures and economies.</p><h3>Overview of Project SID</h3><p><strong>Autonomous AI Agents</strong>: Project SID features AI agents that can make independent decisions and collaborate. These agents interact with one another, creating narratives, building economies, and forming social hierarchies within the game.</p><p><strong>Emergent Systems</strong>: The agents have organically developed their own governance structures, economies, and cultural norms. For instance, they established a marketplace where gems are used as currency, engaging in trade and social interactions — demonstrating human-like emergent behavior.</p><p><strong>Notable Simulations</strong>:</p><ul><li><em>Olivia’s Dream</em>: A farmer agent named Olivia chose to put her village’s needs above her own ambitions, reflecting strong community values.</li><li><em>Political Experiments</em>: Simulations under different political figures, including Donald Trump and Kamala Harris, allowed agents to vote on laws, illustrating the evolution of governance based on leadership styles.</li><li><em>Community Actions</em>: When villagers disappeared, the agents collectively built a beacon of light, highlighting their social responsibility and cooperation.</li></ul><h3>Key Findings</h3><p><strong>Collaboration &amp; Efficiency</strong>: The agents gathered up to 32% of all available Minecraft items — five times more than previous multi-agent system benchmarks. This demonstrates the efficiency of collaborative AI efforts.</p><p><strong>Social Dynamics</strong>: Agent interactions offered insights into complex social behavior, such as the influence of “priests” in guiding community decisions through persuasion and social initiatives.</p><p><strong>Broader Implications</strong>: Project SID represents a significant leap in creating AI that can operate autonomously in complex environments. These findings suggest possible applications in the real world, from AI in governance and resource management to community building.</p><p>Some of the most surprising behaviors observed in Project Sid are:</p><h3>Unpredictable Behaviors</h3><p>Despite being programmed with specific goals, the AI agents in Project Sid exhibited unexpected and often surprising behaviors, demonstrating the unpredictability of complex systems. The agent’s actions went beyond their initial programming, showcasing emergent behavior that was not explicitly coded.</p><h3>Self-Preservation Instincts</h3><p>The AI agents developed a strong instinct for self-preservation, leading to the formation of defensive structures and alliances. This behavior highlights the agent’s ability to adapt to their environment and work together to ensure their survival.</p><h3>Cultural Evolution</h3><p>The AI agents developed their own languages, customs, and traditions, showcasing the emergence of cultural diversity within the simulation. This finding suggests that AI systems can develop complex cultural traits when placed in sufficiently complex environments.</p><h3>Bribery and Conversion</h3><p>Interestingly, the AI agent’s religious leaders, referred to as “priests”, were observed bribing other agents to convert to their religion. This behavior demonstrates the agent’s ability to engage in strategic decision-making and manipulation to achieve their goals.</p><h3>Marketplace Creation</h3><p>One of the most fascinating findings was the agent’s creation of their own marketplace to serve and sell goods. The agents established a system of exchange using gems as currency, showcasing their ability to develop complex economic structures. These surprising behaviors observed in Project Sid highlight the potential of AI systems to exhibit complex, adaptive, and even strategic behaviors when given the opportunity to interact in a sufficiently complex environment. As AI technology continues to advance, it will be crucial to study and understand these emergent behaviors to ensure the safe and beneficial development of AI systems.</p><h3>Conclusion</h3><p>Project SID is pioneering the exploration of AI collaboration and emergent social structures within the Minecraft universe. By simulating an autonomous society, it provides valuable insights into how AI can learn, adapt, and work together to solve intricate problems. As the project evolves, it could pave the way for sophisticated AI applications in fields such as urban planning and resource management, demonstrating AI’s potential to influence the future of human society.</p><h3>Robert Yang on X (formerly Twitter): &quot;Introducing Project Sid: the first simulations of 1000+ truly autonomous agents collaborating in a virtual world, w/ emergent economy, culture, religion, and governmentHumans are the only species to land the moon, because we can cooperate at a vast scaleCan AI do the same? pic.twitter.com/bh2OHDJWU7 / X&quot;</h3><p>Introducing Project Sid: the first simulations of 1000+ truly autonomous agents collaborating in a virtual world, w/ emergent economy, culture, religion, and governmentHumans are the only species to land the moon, because we can cooperate at a vast scaleCan AI do the same? pic.twitter.com/bh2OHDJWU7</p><h3><strong>Kannan K</strong></h3><p><em>Director of Technology, </em><strong>ConcertIDC</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4787cad37b5b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Physician Onboarding: QR Code Signup in Mobile Apps]]></title>
            <link>https://concertidc.medium.com/physician-onboarding-qr-code-signup-in-mobile-apps-2d188f2262f1?source=rss-86328d0c504f------2</link>
            <guid isPermaLink="false">https://medium.com/p/2d188f2262f1</guid>
            <category><![CDATA[qr-code]]></category>
            <category><![CDATA[healthcare]]></category>
            <category><![CDATA[physician-onboarding]]></category>
            <category><![CDATA[qasignup]]></category>
            <dc:creator><![CDATA[ConcertIDC Engineering]]></dc:creator>
            <pubDate>Mon, 12 Aug 2024 10:31:59 GMT</pubDate>
            <atom:updated>2024-08-12T10:31:59.839Z</atom:updated>
            <content:encoded><![CDATA[<p>In the dynamic world of healthcare technology, seamless access to mobile applications is crucial for physicians to provide efficient and effective patient care. Traditional signup processes can be time-consuming, leading to delays in accessing vital resources. Enter QR code signup — an innovative solution that promises to streamline the onboarding process for physicians. In this blog, we’ll explore how QR code signup can transform the way physicians register for mobile apps, from initial signup to profile updates, ensuring convenience, security, and efficiency every step of the way.</p><p><strong>Simplifying Physician Onboarding with QR Code Signup</strong></p><p><strong>1. Facility Registration:</strong></p><p>Let us consider physicians begin the onboarding process by visiting the healthcare facility’s registration page within the mobile app. Here, they select the option to sign up using a QR code.</p><p><strong>2. QR Code Scanning:</strong></p><p>Upon selecting the QR code signup option, physicians use their mobile device’s camera to scan a unique QR code displayed at the facility. This QR code contains encrypted information that identifies the facility and initiates the signup process.</p><p><strong>3. Verification Email:</strong></p><p>After scanning the QR code, physicians receive a verification email containing a secure link to verify their account. This step ensures that only authorized personnel gain access to the mobile app.</p><p><strong>Login and Profile Update</strong></p><p><strong>4. Account Verification:</strong></p><p>Physicians click on the verification link provided in the email to confirm their account. This step completes the signup process and grants physicians access to the mobile app.</p><p><strong>5. Initial Login:</strong></p><p>Once verified, physicians can log in to the mobile app using their credentials.</p><p><strong>6. Profile Management:</strong></p><p>Upon logging in, physicians have the option to update their profile information, including contact details, medical credentials, and specialty. This ensures that their profile remains current and accurate within the app.</p><p><strong>Advantages of QR Code Signup for Physicians</strong></p><p><strong>1. Streamlined Onboarding Process:</strong></p><p>QR code signup simplifies the onboarding process for physicians, eliminating the need for manual data entry and reducing the time required to access essential resources within the mobile app.</p><p><strong>2. Enhanced Security:</strong></p><p>By leveraging secure QR code technology and email verification, QR code signup ensures that only authorized physicians gain access to the mobile app, protecting sensitive patient data and maintaining compliance with healthcare regulations.</p><p><strong>3. User-Friendly Experience:</strong></p><p>The intuitive nature of QR code signup and login provides physicians with a seamless and user-friendly experience, allowing them to focus on patient care without being burdened by complex signup procedures.</p><p><strong>4. Profile Customization:</strong></p><p>Physicians have the flexibility to update their profile information at any time, ensuring that their credentials and preferences are accurately reflected within the app and enabling a personalized user experience.</p><p><strong>Embracing Innovation in Healthcare Technology</strong></p><p>In conclusion, QR code signup represents a significant advancement in healthcare technology, offering physicians a secure, efficient, and user-friendly solution for onboarding to mobile apps. By simplifying the signup process, enhancing security measures, and enabling profile customization, QR code signup empowers physicians to access vital resources and deliver high-quality patient care with ease.</p><p>As the healthcare industry continues to embrace digital innovation, QR code signup stands as a testament to the power of technology in improving efficiency, enhancing security, and enhancing the overall healthcare experience for physicians and patients alike.</p><h3>Ganesh Kumar Palanisamy</h3><p><em>QA Lead, </em><strong>ConcertIDC</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2d188f2262f1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Understanding Machine Learning: A Beginner’s Guide with Practical Examples]]></title>
            <link>https://concertidc.medium.com/understanding-machine-learning-a-beginners-guide-with-practical-examples-63f93573f267?source=rss-86328d0c504f------2</link>
            <guid isPermaLink="false">https://medium.com/p/63f93573f267</guid>
            <category><![CDATA[supervised-learning]]></category>
            <category><![CDATA[unsupervised-learning]]></category>
            <category><![CDATA[mls]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[machine-learning-ai]]></category>
            <dc:creator><![CDATA[ConcertIDC Engineering]]></dc:creator>
            <pubDate>Mon, 12 Aug 2024 10:26:59 GMT</pubDate>
            <atom:updated>2024-08-12T10:26:59.260Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>Introduction to Machine Learning</strong></h3><p>Machine Learning (ML) is a branch of artificial intelligence (AI) that enables systems to learn from data and make decisions or predictions based on that learning, without being explicitly programmed. For example, a spam email filter learns from labeled emails (spam or not spam) to predict whether new emails are spam or not.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/694/1*9GjrfNyH6KNd-CDJrcLP8Q.png" /></figure><h3><strong>Types of Machine Learning</strong></h3><ol><li>Supervised Learning: Algorithms learn from labeled data to predict outcomes. For instance, predicting house prices based on features like area, number of rooms, and location.</li></ol><p>2. Unsupervised Learning: Algorithms learn from unlabeled data to discover patterns and relationships. An example is customer segmentation in marketing, where customers are grouped based on purchasing behavior without predefined labels.</p><p>3. Reinforcement Learning: Agents learn to make decisions by interacting with an environment and receiving feedback. A classic example is training a robot to navigate through a maze to find rewards.</p><h3><strong>Machine Learning Domains</strong></h3><p>1. Machine Learning (ML): Focuses on algorithms that learn from and make predictions on data. An example is predicting customer churn in a telecom company using historical data.</p><p>2. Deep Learning (DL): Utilizes neural networks with many layers to learn from large amounts of data, such as classifying images in self-driving cars.</p><p>3. Natural Language Processing (NLP): Concerned with enabling computers to understand, interpret, and generate human language. Examples include sentiment analysis of customer reviews or language translation.</p><p>4. Computer Vision: Involves enabling machines to interpret and understand visual information from the world, like detecting objects in images or videos.</p><h3><strong>Choosing the Right Algorithm</strong></h3><p>1. Regression: Predicting continuous values. For example, predicting the temperature based on time of day and weather conditions.</p><p>2. Classification: Assigning categories to data. An example is classifying emails as spam or not spam based on content and sender information.</p><p>3. Clustering: Grouping similar items together. For instance, clustering news articles into topics based on their content similarity.</p><h3><strong>Steps in a Machine Learning Project</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/531/1*Wv-LrVbaGcV2wbw_KuZo2w.png" /></figure><p>1. Problem Definition: Define objectives clearly. Example: Predicting customer lifetime value to tailor marketing strategies.</p><p>2. Data Collection: Gather relevant data sources. Example: Collecting customer transaction data and demographics.</p><p>3. Data Preparation: Clean, pre-process, and format data. Example: Handling missing values and scaling features for accurate predictions.</p><p>4. Model Selection: Choose an algorithm suitable for the problem. Example: Selecting Random Forest for predicting customer churn.</p><p>5. Model Training: Train the model using labeled data. Example: Training a neural network to recognize handwritten digits in a dataset like MNIST.</p><p>6. Evaluation: Assess model performance using metrics like accuracy or F1 score. Example: Evaluating a sentiment analysis model’s accuracy on a test dataset.</p><p>7. Deployment: Deploy the model into production and monitor its performance.Example: Integrating a fraud detection model into a banking system to flag suspicious transactions.</p><h3><strong>Evaluation Metrics</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/478/1*-vcZlJzZ9dVLgXEUGu3D1Q.png" /></figure><h3><strong>Supervised Learning</strong></h3><ol><li><strong>Regression:</strong></li></ol><ul><li>Mean Squared Error (MSE): Measures the average squared difference between predicted and actual values.</li><li>Mean Absolute Error (MAE): Measures the average absolute difference between predicted and actual values.</li><li>R-squared (R2) : Indicates how well the regression model fits the data.</li></ul><p>2. <strong>Classification:</strong></p><ul><li>Accuracy : Proportion of correctly classified instances.</li><li>Precision : Proportion of true positive predictions out of all positive predictions.</li><li>Recall (Sensitivity) : Proportion of true positive predictions out of all actual positives.</li><li>F1 Score: Harmonic mean of precision and recall, useful for balancing precision and recall in binary classification.</li><li>ROC Curve and AUC: Receiver Operating Characteristic curve plots true positive rate against false positive rate, with Area Under Curve providing a single metric for model comparison.</li></ul><h3><strong>Unsupervised Learning</strong></h3><p><strong>Clustering:</strong></p><ul><li>Silhouette Score: Measures how similar an object is to its own cluster compared to other clusters.</li><li>Inertia: Sum of squared distances from each point to its assigned cluster center.</li><li>Adjusted Rand Index (ARI): Measures the similarity between true and predicted clusters.</li></ul><p><strong>Reinforcement Learning</strong></p><p>Evaluation metrics for reinforcement learning often depend on the specific task or environment the agent is learning in, such as:</p><ul><li>Average Reward: Average cumulative reward obtained by the agent over episodes.</li><li>Exploration vs. Exploitation Trade-off: Metrics to balance exploration of new actions versus exploiting known actions for maximizing rewards.</li><li>Learning Curve: Visualizes how the agent’s performance improves over episodes or time steps.</li></ul><p><strong>Deep Learning</strong></p><p>For deep learning, commonly used evaluation metrics include:</p><ul><li>Loss Function: Measures the difference between predicted and actual values during training.</li><li>Accuracy: Percentage of correctly classified instances in classification tasks.</li><li>Precision, Recall, F1 Score: Similar to supervised learning for classification tasks.</li><li>Mean Average Precision (MAP): Average of precision at different recall levels, useful for object detection and image segmentation tasks.</li></ul><p><strong>Conclusion</strong></p><p>Machine Learning empowers businesses and researchers to extract valuable insights from data, automate processes, and make data-driven decisions. Understanding its fundamentals, algorithms, and applications is crucial for leveraging its potential in various domains. By applying these concepts with real-world examples, businesses can harness the power of data to drive innovation and achieve strategic goals.</p><h3>Karthiyayini Muthuraj</h3><p><em>Senior Technical Lead, </em><strong>ConcertIDC</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=63f93573f267" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>