<?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 Akshay Sunderwani on Medium]]></title>
        <description><![CDATA[Stories by Akshay Sunderwani on Medium]]></description>
        <link>https://medium.com/@akshya672222?source=rss-2cfa25d159d3------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*aEI4m-V4NlJYk94C.jpg</url>
            <title>Stories by Akshay Sunderwani on Medium</title>
            <link>https://medium.com/@akshya672222?source=rss-2cfa25d159d3------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 14:12:11 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@akshya672222/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[How I Refactored My Resume Like a Codebase — and Turned It Into a Senior-Level Story]]></title>
            <link>https://medium.com/@akshya672222/how-i-refactored-my-resume-like-a-codebase-and-turned-it-into-a-senior-level-story-c458f66de2a5?source=rss-2cfa25d159d3------2</link>
            <guid isPermaLink="false">https://medium.com/p/c458f66de2a5</guid>
            <category><![CDATA[self-improvement]]></category>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Akshay Sunderwani]]></dc:creator>
            <pubDate>Fri, 20 Feb 2026 00:54:17 GMT</pubDate>
            <atom:updated>2026-02-20T00:54:17.973Z</atom:updated>
            <content:encoded><![CDATA[<p>Most engineers treat their resume like legacy code — it works, so don’t touch it. I was the same way. After 9+ years across Microsoft, AWS, Boxed, and shipping my own App Store product, my resume was still a cobbled-together Word doc that undersold almost everything I’d done.</p><p>Then I decided to audit it the way I’d audit a codebase: systematically, with metrics, and without mercy.</p><p>Here’s what I found, what I changed, and the Typst template I built and open-sourced so you can do the same.</p><p>But first — here’s the before and after:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9gVrlePPD7ZQG9Ue5B6l6g.png" /><figcaption>Before: 2-page LaTeX resume with verbose bullets, mentoring as a job entry, and 80% test coverage buried in the middle</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZoKIzx7Bojua6XUDwa_wiA.png" /><figcaption>After: Custom Typst template with summary box, skill chips, and result-first bullets</figcaption></figure><p>Same person. Same experience. Completely different story.</p><p>Now let me break down exactly what changed and why.</p><h3>The Audit: Rating My Own Resume</h3><p>I laid my resume next to everything I’d actually shipped and rated it honestly. I realized I’d shipped real impact but documented none of it and my resume looked mid even though my work wasn’t.</p><p>The content was there, but the presentation was wrong. Here’s what was broken:</p><ol><li><strong>Activity-first bullets that said nothing.</strong></li></ol><p>My Microsoft section opened with “Led the development and delivery of key community management features.” Led what? Delivered what? There’s no metric, no outcome, no reason for a hiring manager to keep reading.</p><p><strong>2. Mentoring listed as a job title.</strong></p><p>I had “ADPList.org/Topmate.io — Mentor” as a standalone role with four verbose bullets about “delivering actionable feedback” and “empowering mentees to gain confidence.” Mentoring is valuable, but giving it a full Experience entry pushed real engineering roles off the first page and diluted the signal.</p><p><strong>3. Vague filler bullets that could be anyone’s.</strong></p><p>My Boxed section had “Enhanced user experience and accessibility, ensuring a more inclusive and user-friendly platform.” That sentence applies to every engineer at every company. It’s empty calories.</p><p><strong>4. Two pages of padding.</strong></p><p>The resume ran two full pages. It included a Teaching Assistant role from 2017, a “Leadership: Coaching and Mentoring upcoming students” skill category, and a line about being recognized as “Mr. Professional.” None of these told a senior engineering story. They just took up space.</p><p><strong>5. Key achievements buried or missing.</strong></p><p>My strongest work — a ~70% rendering performance improvement at Microsoft, a DAU growth from 15K to 75K — wasn’t even on the resume. The Microsoft section led with “Led the development and delivery” and peaked at “achieving over 80% code coverage.” Coverage is table stakes, not a headline.</p><p>6. <strong>No mention of my shipped product.</strong></p><p>I’d founded and shipped <a href="https://apps.apple.com/us/app/parentpal-storywriter/id6753812370">ParentPal StoryWriter </a>— an AI-powered iOS app live on the App Store with a 14-module SwiftPM architecture. It wasn’t anywhere on the resume.</p><h3>The Rewrite: Seven Specific Changes</h3><p>I didn’t rewrite everything. I made seven targeted changes, each with a clear rationale.</p><h4>1. Flip every bullet to result-first</h4><p><strong>Before:</strong> “Led the development and delivery of key community management features for Microsoft Teams — Communities, significantly driving user engagement and enhancing platform functionality.”</p><p><strong>After:</strong> “Reduced cell rendering latency by ~70% (536ms → 156ms) via Instruments profiling; followed up with a ~300-file refactor eliminating systemic double-render cycles — shipped with zero regressions.”</p><p>The old bullet described an activity. The new one opens with a measurable outcome that any engineering manager immediately understands.</p><h4>2. Demote mentoring from a role to a line</h4><p><strong>Before:</strong> “A full “ADPList.org/Topmate.io — Mentor” entry with four bullets taking up a quarter of the first page.”</p><p><strong>After</strong>:<strong> “</strong>One line under Leadership &amp; Education: “Top 1% mentor on ADPList/Topmate with 50+ sessions; helped mentees secure roles at Apple, Google, AWS.<strong>”</strong></p><p>The impact is preserved. The space is freed for engineering work.</p><h4>3. Replace vague bullets with specific features</h4><p><strong>Before</strong>:<strong> </strong>“Enhanced user experience and accessibility, ensuring a more inclusive and user-friendly platform for all users.” (Boxed)</p><p><strong>After</strong>:<strong> </strong>“Increased average order value by designing and A/B testing high-impact features (Smart Stock Up, Cart Builder) across both iOS and Android.”</p><p>One is a platitude. The other names the features, the method (A/B testing), and the business outcome.</p><h4>4. Cut from two pages to one</h4><p>I removed the Teaching Assistant role, the “Leadership” skill category, the “Mr. Professional” recognition, and four mentoring bullets. I trimmed every remaining bullet to one line where possible. The resume went from two pages of noise to one page of signal.</p><h4>5. Lead with your strongest metric</h4><p>The old Microsoft section opened with community features. I reordered so the first thing you read is a 70% latency reduction with specific numbers (536ms → 156ms), followed by the ~300-file refactor. I then added metrics that weren’t on the old resume at all: DAU growth from 15K to 75K, 96%+ test coverage (up from 80%), and the Microsoft Hackathon placement.</p><h4>6. Add the shipped product</h4><p>ParentPal StoryWriter got its own section: Founder &amp; iOS Engineer. I described the 14-module SwiftPM architecture, the AI content safety pipeline, StoreKit 2 subscriptions, and the offline narration system. This showed end-to-end product ownership — something the old resume completely lacked.</p><h4>7. Reframe the opening from generic to specific</h4><p><strong>Before:</strong> “Seasoned iOS and mobile application developer with over 7 years of experience in designing and deploying scalable, high-performance applications.”</p><p><strong>After: </strong>“Senior iOS engineer with 9+ years of experience designing and shipping scalable, high-performance mobile applications using Swift, SwiftUI, Objective-C, and Kotlin. Led cross-platform features for Microsoft Teams with 96%+ test coverage, ~70% rendering performance improvement, and high post-release reliability. Founded ParentPal StoryWriter — an AI-powered iOS app live on the App Store.”</p><p>The old summary was a template sentence. The new one is a pitch with three proof points.</p><h3>The Template: Why Typst, Not LaTeX or Word</h3><p>After fixing the content, I rebuilt the presentation layer. I chose <a href="https://typst.app/">Typst</a> over LaTeX for the same reason I’d choose SwiftUI over UIKit for a new project — it’s modern, readable, and compiles in milliseconds instead of seconds.</p><p>The template has:</p><p><strong>Summary box</strong> with a left-border accent — your pitch in 3 sentences</p><p>Skill chips<strong> — themed tag-style badges that look clean without taking much space</strong></p><p>Full-width section headers<strong> — colored bars that create clear visual hierarchy</strong></p><p>One theme color<strong> — change one rgb() value and everything updates</strong></p><p>The biggest challenge was fitting everything on one page. I went through multiple iterations — adjusting margins, leading, list spacing, and skill chip padding until the content filled the page evenly without cramping or spilling. It’s the same iterative process as performance optimization: measure, adjust, measure again.</p><p><strong>The template is open-source.</strong> Fork it, replace the placeholder content, run ./build.sh, and you have a PDF.</p><h3>The Result</h3><p>The same experience, reframed:</p><p>Every bullet leads with a result, not an activity. Metrics are specific but not absolute. Architecture and impact replace volume and vanity. The visual design is clean, professional, and fits on exactly one page.</p><p>The resume went from cluttered and underselling to sharp, focused, and aligned with the level of work I’ve actually done. It’s still not perfect — a resume is never finished. It evolves with every role, every shipped feature, and every lesson learned.</p><h3>What I’d Tell Other Engineers</h3><p><strong>Audit your resume against your actual work, not your memory of it.</strong> I found three missing roles and a dozen undersold achievements by going through my projects one by one. Your memory compresses things. The evidence doesn’t.</p><p>Write result-first, not activity-first.<strong> Start every bullet with the outcome. “Reduced latency by 70%” hits harder than “Worked on performance optimization.”</strong></p><p>Kill your darlings.<strong> That commit count you’re proud of? If it doesn’t tell a senior-level story, cut it. Replace it with something that shows judgment, not just output.</strong></p><p>Invest in the presentation layer.<strong> Your resume is a product. It has users (recruiters, hiring managers), it has a UX (scannability, hierarchy), and it ships (as a PDF). Treat it accordingly.</strong></p><p><em>I used Claude as a sparring partner through this process — to pressure-test my bullets, catch vague language, and iterate on the Typst template layout. The same way I’d use Instruments to profile a performance issue, I used an AI tool to profile my resume. The judgments — what to cut, what to lead with, how to frame 9 years of work — were mine. The tool just made the feedback loop faster.</em></p><p><em>The Typst template is MIT-licensed and available on </em><a href="https://github.com/akshya672222/typst-resume-template"><em>GitHub</em></a><em>. Fork it, make it yours.</em></p><p>If you found this useful, I mentor engineers on resume writing, interview prep, and career growth through <a href="https://topmate.io/akshay_sunderwani">Topmate</a> and <a href="https://adplist.org/mentors/akshay-sunderwani">ADPList</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c458f66de2a5" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why Your AI Content Automation Sounds Fake — And How I Fixed It]]></title>
            <link>https://medium.com/@akshya672222/why-your-ai-content-automation-sounds-fake-and-how-i-fixed-it-692676965e0b?source=rss-2cfa25d159d3------2</link>
            <guid isPermaLink="false">https://medium.com/p/692676965e0b</guid>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Akshay Sunderwani]]></dc:creator>
            <pubDate>Tue, 17 Feb 2026 23:50:01 GMT</pubDate>
            <atom:updated>2026-02-17T23:50:01.730Z</atom:updated>
            <content:encoded><![CDATA[<p>From fabrication prevention to production error handling — lessons learned from automating my entire social media workflow</p><p>The Problem: Content Creation at Scale</p><p>As a senior iOS engineer at Microsoft Teams, I barely have time to breathe, let alone maintain a consistent social media presence. But building a personal brand as a technical expert and mentor requires regular, high-quality content across LinkedIn, Medium, and X.</p><p>The traditional approach — manually writing posts, copying content between platforms, remembering to post consistently — was unsustainable. I needed automation, but not the generic, robotic kind. I needed intelligent automation that could maintain my voice, technical depth, and authenticity.</p><p>So I built an end-to-end content automation pipeline using Claude API. This isn’t theoretical — it’s the system generating this very article you’re reading.</p><p>The Architecture: More Than Just “AI Writes Posts”</p><p>Most people think AI content automation is just “ask ChatGPT to write a post.” That approach produces generic, obviously AI-generated content that damages your brand more than it helps.</p><p>My pipeline is built around three core principles:</p><p>1. Context Preservation: The system knows everything about my background, experience, and voice.<br>2. Platform Optimization: Different content formats for LinkedIn (750 words), Medium (1500+ words), and X (280 chars).<br>3. Quality Gates: Multiple validation layers ensure authenticity and accuracy.</p><p>The System Prompt: The Foundation of Everything</p><p>The system prompt is where the magic happens. This isn’t a simple “write like Akshay” instruction — it’s a comprehensive specification covering a verified profile (2,000+ words of verified facts), fabrication prevention rules, voice guidelines, and platform-specific format requirements.</p><p>The most critical section deals with fabrication prevention:</p><p>CRITICAL RULE — NO FABRICATION:<br>- You MUST only use facts, experiences, and details that appear in the verified profile<br>- NEVER invent stories, anecdotes, conversations, or scenarios that didn’t happen<br>- NEVER fabricate job offers, rejections, or career events<br>- NEVER make up metrics, numbers, or outcomes not in the profile</p><p>This rule exists because I learned the hard way that AI will confidently invent plausible-sounding details. In early iterations, Claude would create stories about “a mentee who landed at Google” or “when I was building feature X at Microsoft.” These sound authentic but are complete fabrications that could damage credibility.</p><p>What failure looked like vs. the fix:</p><p>Early output (fabricated): “Last year, I mentored a junior dev who went on to land a Staff Engineer role at Google — proof that consistent guidance compounds over time.”</p><p>After fabrication prevention rules: “As someone who’s navigated the Microsoft career ladder from IC to Senior, I’ve seen firsthand how systems thinking separates good engineers from great ones.”</p><p>The difference is subtle but critical. The first sounds like a LinkedIn cliche. The second is grounded, specific, and actually mine.</p><p>Lessons from Building ParentPal’s AI Validation Pipeline</p><p>The content automation pipeline wasn’t my first AI integration project. While building ParentPal StoryWriter, I implemented an AI validation system for generated stories that taught me crucial lessons about production AI systems.</p><p>One critical lesson: unbounded recursion will kill you in production. Always use a depth guard — cap repair cycles at 3 attempts before flagging for human review. Without it, a persistently bad AI response will recurse until you hit a stack overflow or burn through your API budget.</p><p>The full Swift implementation with DocC comments is in my GitHub repo: <a href="https://github.com/akshya672222/content-automation-pipeline">https://github.com/akshya672222/content-automation-pipeline</a></p><p>On-Device vs Cloud AI: The Microsoft Hackathon Experience</p><p>At Microsoft Hackathon 2025, my team placed 4th with an on-device auto-moderation prototype for spam detection. This experience highlighted crucial tradeoffs between on-device and cloud AI that directly influenced my automation pipeline design.</p><p>On-device advantages: zero latency for real-time validation, complete privacy, no API costs at scale. Cloud AI advantages: access to latest models, no device resource constraints, easier updates.</p><p>For content automation, I chose cloud AI (Claude) because content generation needs the most advanced reasoning abilities, and personal use volume makes API costs negligible vs development time. However, I’m monitoring Apple’s on-device AI improvements closely.</p><p>The Publishing Layer: Playwright and Platform APIs</p><p>LinkedIn doesn’t have a public posting API for personal profiles. The approach is browser automation via Playwright — navigating to the feed, authenticating, and triggering the post creation flow programmatically.</p><p>I’m intentionally not publishing specific CSS selectors here. LinkedIn’s UI ships updates constantly and hardcoded selectors in a public article will be broken within weeks. The full implementation with up-to-date selectors lives in my GitHub repo: <a href="https://github.com/akshya672222/content-automation-pipeline">https://github.com/akshya672222/content-automation-pipeline</a></p><p>Key pattern: abstract selectors into named helpers so when LinkedIn breaks something, you fix it in one place.</p><p>Performance Optimization and Cost Management</p><p>Prompt Caching: Anthropic’s prompt caching reduces costs by ~50% since my system prompt rarely changes.</p><p>Concurrent Generation: asyncio.gather to generate content for all platforms in parallel — cuts total generation time significantly.</p><p>Error Recovery: The pipeline stores partial results and resumes from failures. If LinkedIn posting fails but Medium succeeds, it only retries the failed platform.</p><p>Full code samples for all patterns are in the GitHub repo.</p><p>What I’d Do Differently</p><p>1. Content Scheduling: Currently I trigger generation manually. True automation requires scheduling.<br>2. Performance Analytics: Connecting platform APIs for engagement metrics would close the feedback loop.<br>3. A/B Testing: Multiple variants per platform + testing would optimize performance over time.<br>4. Voice Evolution: Updating voice guidelines based on recent high-performing content would keep output authentic.</p><p>Conclusion: The Metric That Actually Changed</p><p>Six months in, the most meaningful outcome isn’t time saved — it’s consistency. Before this pipeline, I’d post in bursts when I had energy, then go dark for weeks. The system enforces a cadence my willpower couldn’t.</p><p>The fabrication prevention work turned out to be the most important engineering decision. It’s what separates a pipeline that builds a brand from one that quietly destroys it. Every post that goes out is grounded in something I actually did, thought, or built.</p><p>If you’re starting this journey: solve fabrication first, automate second. Everything else is optimization.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=692676965e0b" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>