<?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 Julius Uy on Medium]]></title>
        <description><![CDATA[Stories by Julius Uy on Medium]]></description>
        <link>https://medium.com/@julius.uy?source=rss-896373bd920f------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*VEqMvkX8ZEto5DK-vGaeYg.jpeg</url>
            <title>Stories by Julius Uy on Medium</title>
            <link>https://medium.com/@julius.uy?source=rss-896373bd920f------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 21:26:34 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@julius.uy/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[Android 17 and the VVC Video Compression Standard]]></title>
            <link>https://medium.com/@julius.uy/android-17-and-the-vvc-video-compression-standard-2d55c1a8023d?source=rss-896373bd920f------2</link>
            <guid isPermaLink="false">https://medium.com/p/2d55c1a8023d</guid>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[mobile-app-development]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[android]]></category>
            <dc:creator><![CDATA[Julius Uy]]></dc:creator>
            <pubDate>Sun, 01 Mar 2026 11:52:22 GMT</pubDate>
            <atom:updated>2026-03-01T11:52:22.692Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Zm3Ss9jIANZzkDuryyn1IQ.png" /></figure><p>If you’ve spent any meaningful amount of time in the trenches of a video streaming service, you know that video delivery is a ruthless game of inches. You spend weeks tweaking your video QoS configuration just to squeeze an extra few percent of visual fidelity out of a sub-megapixel stream. When it comes to video streaming, lowering CDN bills while making the user experience so delightfully seamless is a perpetual priority.</p><p>The quiet arrival of official <strong>VVC (Versatile Video Coding)</strong> support in Android 17 brings a new player in the long game of video compression. So, let’s talk about VVC. What it is, the math that makes it tick, and why your backend and mobile teams should be paying close attention.</p><h3>The Codec Wars: A File Size Story</h3><p>To understand why VVC is a big deal, we have to look at the blood-stained battlefield of video compression. If we use a standard 4K high-motion video file as our baseline, here is how the math roughly plays out when we optimize for the <em>exact same</em> subjective visual quality:</p><ul><li><strong>H.264 (AVC):</strong> The beloved grandfather. Let’s say our baseline 4K video file size is <strong>1 GB</strong>. It works on everything from your grand-aunt’s decade-old tablet to a smart toaster, but it eats bandwidth for breakfast.</li><li><strong>H.265 (HEVC):</strong> The heavy lifter. It brought that file size down to <strong>~500 MB</strong> (a 50% reduction). It made 4K viable, but the licensing was a bureaucratic nightmare that gave lawyers yachts. VP9 was free, but adoption was slow.</li><li><strong>AV1:</strong> The open-source darling. It brought the file size down further to <strong>~350 MB</strong> (about 30% better than HEVC). We love AV1. It’s royalty-free. But decoding it on older hardware without a dedicated chip? Watch your battery plummet fast enough to make you scream, <em>“Susan!!! The remote is out of battery!”</em></li><li><strong>VVC (H.266):</strong> The new king. It brings that file size down to <strong>~250 MB</strong> (another 30% to 50% reduction over HEVC, and often 15–20% smaller than AV1).</li></ul><p>But VVC isn’t just about shrinking file sizes. As the name suggests, it is <em>Versatile</em>. It is mathematically designed not just for Hollywood movies, but for the chaotic mix of 8K streaming, 360-degree VR environments, and crisp cloud gaming where text needs to be highly legible.</p><h3>Bridging the Gap: Visualizing ‘Versatile’</h3><p>Traditional videocompression often struggles with mixed-complexity content. A codec optimized for a slow-moving drama might fail spectacularly on high-motion sports or complex, static textures (like text). VVC handles this by being highly adaptive.</p><p>Think of it this way:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*p5EFsZX_itwYDZvCmgnvDw.png" /><figcaption>VVC encoding evaluates raw videos and identifies motion density and applies corresponding entropies to have the series of images remain faithful to the naked eye.</figcaption></figure><p>VVC’s ‘Versatile’ approach means it analyzes a video frame (in this case, a complex soccer match) and intelligently breaks it into different Coding Units (CUs). A static, textured stadium crowd gets larger, simpler blocks. The explosive action around the players gets tiny, complex blocks, dynamically adjusting to motion and texture simultaneously. The process of encoding this complexity into a stream and <em>decoding</em> it back in real-time is what Android 17’s new integration facilitates.</p><h3>The Technical Underbelly: Why VVC is so Efficient</h3><p>If we already have AV1, why do we need VVC? The efficiency doesn’t come from magic. It comes from significantly more complex mathematics that will happily fry an unoptimized CPU.</p><p>VVC introduces evolutionary optimizations across the standard hybrid video coding model. The two biggest architectural differences — the ones that will make software engineers nod in solemn appreciation — are in <strong>Partitioning</strong> and <strong>Loop Filtering</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pBGtYRA9_TtQFlRMPM84eA.png" /></figure><h4>1. Partitioning: The QTMT Structure (Or, “Squares are for squares”)</h4><p>In HEVC, frames were split into 64x64 Coding Tree Units (CTUs), partitioned using a rigid <strong>Quadtree</strong> structure. A block could only split into four smaller, equal squares. It was simple, but nature doesn’t operate in perfect squares.</p><p>VVC introduces the <strong>Quadtree with nested Multi-type Tree (QTMT)</strong>. It allows CTUs up to 128x128, and more importantly, allows non-square splits:</p><ul><li><strong>Quadtree (QT):</strong> Your standard square split.</li><li><strong>Binary Tree (BT):</strong> Splitting into two rectangles.</li><li><strong>Ternary Tree (TT):</strong> Splitting into three rectangles (e.g., a 1:2:1 ratio).</li></ul><p>This mathematically expensive Rate-Distortion Optimization (RDO) allows the codec to perfectly trace the true boundaries of an object — like the curved edge of a moving car — rather than awkwardly forcing a grid of squares over it.</p><h4>2. The Loop Filter: ALF (Adaptive Loop Filter)</h4><p>HEVC and AV1 use loop filters inside the inter-prediction loop to smooth block edges. HEVC uses Deblocking (DBF) and Sample Adaptive Offset (SAO). AV1 throws in CDEF and Restoration Filters.</p><p>VVC says, “Hold my beer,” and adds <strong>ALF (Adaptive Loop Filter)</strong>.</p><p>ALF is highly adaptive, using 2D diamond-shaped filters (e.g., 7x7 for luma) applied on a 4x4 block basis. The filter coefficients adapt <em>for each CTU</em> and can be signaled at the frame level. This means VVC is exceptionally good at restoring fine, high-frequency textures (like the pores on an actor’s face or the grain of a wooden table) that simpler filters turn into a blurry mess.</p><h3>Visualizing the Madness</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pBGtYRA9_TtQFlRMPM84eA.png" /></figure><h3>Part 3: What New Features and Applications does Android 17 Enable?</h3><p>Android 17 providing OS-level VVC support is the essential prerequisite for device manufacturers (OEMs) to ship hardware capable of decoding it efficiently. When that hardware arrives, these scenarios become viable:</p><h3>1. Seamless 8K Mobile Streaming</h3><p>HEVC bitrates for 8K (around 80–100 Mbps) are too high for consistent mobile delivery. VVC can deliver comparable 8K experiences in the 40–60 Mbps range, making it viable over robust 5G or Wi-Fi 6.</p><h3>2. High-Fidelity Cloud Gaming and Screen Sharing</h3><p>Screen content is notorious for sharp edges (text) and massive flat color areas. VVC includes specific tools, such as <strong>Intra Block Copy (IBC)</strong> and <strong>Palette Coding</strong>, tailored for screen-captured content. It allows much lower latency and significantly fewer artifacts (chroma bleeding on text) in remote desktop or game streaming scenarios compared to HEVC.</p><h3>3. Native Support for ‘Immersive’ Video</h3><p>VVC has built-in 360-degree video coding tools, handling the projection mapping (like equirectangular projection) <em>inside</em> the codec. Previously, apps had to manage this stitching and projection in app logic or via custom ExoPlayer renderers. Android 17 could now present a natively decoded, immersive video stream to the renderer, massively simplifying development.</p><h3>Conclusion: Preparation for the VVC Era</h3><p>As engineering leaders, we shouldn’t immediately switch our encoding pipelines to VVC. The encoding complexity is still <em>massive</em> (VVC encoding is roughly 10x more computationally expensive than HEVC).</p><p>However, Android 17’s move is the signal to start preparing. If your application handles:</p><ul><li><strong>Ultra-High Definition (4K/8K)</strong></li><li><strong>Screen Casting/Cloud Gaming</strong></li><li><strong>VR/360 Content</strong></li></ul><p>…then you should be monitoring the market for Android 17 devices with <strong>VVC Hardware Acceleration</strong> (e.g., checking MediaCodecInfo.isHardwareAccelerated() for the video/vvc MIME type).</p><p>The efficiency gains are real, and the versatility of this new standard is exactly what the next generation of mobile experiences requires.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2d55c1a8023d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Fortran in Weather and Climate Research: Migration Challenges, Costs, and Strategic Decisions]]></title>
            <link>https://medium.com/@julius.uy/fortran-in-weather-and-climate-research-migration-challenges-costs-and-strategic-decisions-66c985bae4a2?source=rss-896373bd920f------2</link>
            <guid isPermaLink="false">https://medium.com/p/66c985bae4a2</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[python]]></category>
            <category><![CDATA[weather]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[climate-change]]></category>
            <dc:creator><![CDATA[Julius Uy]]></dc:creator>
            <pubDate>Sun, 24 Aug 2025 06:59:20 GMT</pubDate>
            <atom:updated>2025-08-24T07:41:11.582Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wpd7jUS9TJwDBMcZb8o4cQ.png" /></figure><p>The weather and climate research community has always faced a decision point regarding the continued use of Fortran versus migration to modern programming languages like Python. Based on a comprehensive analysis of current usage patterns, technical requirements, and migration costs, the evidence strongly suggests that staying with Fortran while implementing strategic modernization is the most cost-effective approach for most organizations in the near term.</p><h3><strong>Fortran Dominance in Weather and Climate Modeling</strong></h3><p>Fortran maintains an overwhelming presence in weather and climate research, with as much as 80% of major climate models, including the <a href="https://github.com/E3SM-Project/E3SM">Energy Exascale Earth System Model (E3SM)</a>, <a href="https://github.com/wrf-model/WRF">Weather Research and Forecasting Model (WRF)</a>, and <a href="https://github.com/escomp">Earth Science Community Modeling</a> (ESCOMP), written primarily in Fortran. This dominance stems from Fortran’s specific design for numerical computing and its ability to generate highly optimized machine code for mathematical operations.¹</p><p>The major climate modeling centers globally — including NOAA/GFDL, <a href="https://ncar.ucar.edu/">NCAR</a>, and international institutions — continue to rely on Fortran-based systems because these models represent decades of scientific validation, testing, and optimization. The Weather Research and Forecasting Model alone demonstrates Fortran’s continued relevance, recently achieving a 2.08x speedup through OpenMP GPU optimizations while maintaining its Fortran codebase.²</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/360/1*-pKVlvs7Nn5NRUTWYA2swQ.png" /><figcaption>Shut up, Schultz. Imma use Fotran over my dead bodies. Now release the Bismarck</figcaption></figure><h3>Performance Implications and Technical Realities</h3><p>The performance differential between Fortran and Python presents perhaps the most compelling argument for maintaining Fortran systems. Benchmark studies consistently show Fortran executing 30–60 times faster than Python for numerical computing tasks. Even when Python leverages optimized libraries like NumPy (which is itself partially written in Fortran), the performance gap remains substantial for the intensive computations required in climate modeling.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*yWtXMKWgLoubvEaq.png" /><figcaption>Programming Language Job Market Comparison in Scientific Computing</figcaption></figure><p>Modern Fortran compilers excel at optimizing array operations, loop vectorization, and memory access patterns — critical capabilities for climate simulations that process massive datasets and perform complex mathematical operations across multi-dimensional grids. The language’s design specifically restricts pointer usage and aliasing in ways that enable aggressive compiler optimizations, allowing code to “run close to the metal”.²</p><h3>WMO Standards and Compliance Considerations</h3><p>Regarding World Meteorological Organization (WMO) compliance, the programming language choice has no direct impact on standards adherence. WMO standards focus on data quality, measurement accuracy, sensor specifications, and site classifications rather than the computational implementation. Organizations can maintain WMO compliance regardless of whether they use Fortran, Python, or other languages, as long as their systems produce accurate meteorological data within specified uncertainty limits.³ ⁴</p><p>The WMO Technical Regulations emphasize standardization of meteorological observations, data exchange protocols, and forecasting procedures rather than prescribing specific programming languages. This provides flexibility for organizations to choose technical implementations that best serve their operational requirements while maintaining regulatory compliance.⁴</p><h3>Migration Costs and Complexity Analysis</h3><p>The financial and technical costs of migrating from Fortran represent a significant barrier to change. Analysis of migration strategies reveals that full migration to Python would cost approximately $15 million over 3 years for a typical climate modeling organization, while suffering severe performance degradation. The <a href="https://www.osti.gov/biblio/1970284">Los Alamos National Laboratory’s 2023 assessment</a> identified multiple risk factors that make large-scale Fortran migration prohibitively expensive.⁵ ⁶</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UadcKvGXrVCbnLKYxEWCAw.png" /></figure><h4>Technical Challenges</h4><ul><li>Legacy codebases often exceed one million lines of tested, validated code.⁶</li><li>Climate models require bit-for-bit reproducibility for scientific credibility, making any translation risky.⁷</li><li>Existing Fortran codes integrate complex physics parameterizations developed over decades.⁸</li></ul><h4>Human Resource Challenges</h4><ul><li>Attracting and retaining staff for Fortran projects has become increasingly difficult.⁹</li><li>The pool of developers with deep Fortran expertise continues to shrink.⁹</li><li>Training existing staff in new languages requires substantial time investment.¹⁰</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/240/1*lMb6uEwReCJBR8kEneJX2A.png" /><figcaption>At least I can expense this in this month’s claims</figcaption></figure><h3>Skills Landscape and Training Considerations</h3><p>The skills shortage in Fortran programming represents a critical long-term challenge. A 2023 research shows that the job market only has 1,050 Fortran positions compared to 83,591 Python jobs in the United States.⁹ This shortage manifests in several ways:</p><h4>Academic Training Gap</h4><ul><li>Fortran is no longer widely taught in computer science curricula.¹⁰</li><li>New graduate students typically learn Python, R, or Julia rather than Fortran.¹¹</li><li>The disconnect between academic training and operational needs creates recruitment challenges.¹⁰</li></ul><h4>Professional Development</h4><ul><li>Training computational scientists in Fortran requires 6–12 months for basic proficiency.⁹</li><li>Scientists can typically learn Python more quickly due to its intuitive syntax.¹²</li><li>However, learning effective scientific Python programming (including NumPy, SciPy, and performance optimization) also requires significant time investment.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/400/0*V06PsxiuAgW6vcV7.jpg" /><figcaption>Not if I steal from Lewis!</figcaption></figure><h3>Strategic Alternatives and Hybrid Approaches</h3><p>Rather than wholesale migration, several organizations are pursuing hybrid Fortran-Python approaches that leverage the strengths of both languages.¹³ These strategies include:</p><h4>Fortran-Python Interfaces</h4><ul><li>Modern coupling frameworks enable calling Python machine learning codes from Fortran applications.¹³</li><li>Performance overhead typically ranges from 5–15%, which is manageable for many applications.¹³</li><li>This approach allows organizations to incorporate modern AI/ML capabilities while preserving validated Fortran cores.¹³</li></ul><h4>Incremental Modernization</h4><ul><li>Updating Fortran codes to modern standards (Fortran 2008, 2018) rather than changing languages.¹⁴</li><li>Implementing GPU acceleration through OpenMP or OpenACC directives.¹⁵</li><li>Maintaining Fortran computational engines while developing Python interfaces for user interaction.¹²</li></ul><h3>Institutional Inertia and Risk Assessment</h3><p>The climate modeling community exhibits significant institutional inertia that extends beyond technical considerations. Domain experts who have invested years in Fortran-based research have little incentive to change unless new languages offer substantial advantages.¹² This inertia manifests in several ways:</p><h4>Scientific Validation Requirements</h4><ul><li>Climate models undergo extensive validation processes that can take years to complete.²</li><li>Any language change risks introducing bugs or numerical inconsistencies that could invalidate scientific results.⁷</li><li>The reputation and career impacts of failed migrations create strong conservative pressures.¹⁶</li></ul><h4>Funding and Resource Allocation</h4><ul><li>Research funding typically focuses on scientific outcomes rather than software modernization.¹⁶</li><li>Migration projects compete with core research activities for limited resources.⁶</li><li>Justifying migration costs to funding agencies requires demonstrating clear scientific benefits.⁶</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/686/0*QZ7DHV7oyUY_mMM6" /><figcaption>I ain’t going nowhere, Linda. I AIN’T GONNA MOVE AN INCH!!!</figcaption></figure><h3>Conclusion</h3><p>The evidence strongly indicates that premature migration from Fortran would be prohibitively expensive and technically risky for most organizations. The combination of performance requirements, validation constraints, skills shortages, and financial costs makes staying with Fortran — while implementing strategic modernization — the most prudent approach. Organizations should focus resources on enhancing their existing Fortran capabilities rather than pursuing wholesale replacement, at least until more mature and proven alternatives emerge that can match Fortran’s performance characteristics while offering significant additional benefits.</p><p>____</p><p>¹ <a href="https://fortran-lang.discourse.group/t/weather-and-climate-modeling-codes-from-fortran-to-c/1531">https://fortran-lang.discourse.group/t/weather-and-climate-modeling-codes-from-fortran-to-c/1531</a></p><p>² <a href="https://partee.io/2021/02/21/climate-model-response/">https://partee.io/2021/02/21/climate-model-response/</a></p><p>³ <a href="https://www.hypermeteo.com/wp-content/uploads/2021/12/3-Guide-to-the-design-enhancement-compliance-and-certification-of-on-site-meteorological-and-climatological-monitoring-stations-and-networks.pdf">https://www.hypermeteo.com/wp-content/uploads/2021/12/3-Guide-to-the-design-enhancement-compliance-and-certification-of-on-site-meteorological-and-climatological-monitoring-stations-and-networks.pdf</a></p><p>⁴ <a href="https://wmo.int/about-us/governance/technical-commissions/standards-and-recommended-practices">https://wmo.int/about-us/governance/technical-commissions/standards-and-recommended-practices</a></p><p>⁵ <a href="https://www.osti.gov/biblio/1970284">https://www.osti.gov/biblio/1970284</a></p><p>⁶ <a href="https://fortran-lang.discourse.group/t/cost-of-modernising-or-rewriting-fortran-codebase-s/9138">https://fortran-lang.discourse.group/t/cost-of-modernising-or-rewriting-fortran-codebase-s/9138</a></p><p>⁷ <a href="https://gmd.copernicus.org/preprints/gmd-2020-18/gmd-2020-18-manuscript-version6.pdf">https://gmd.copernicus.org/preprints/gmd-2020-18/gmd-2020-18-manuscript-version6.pdf</a></p><p>⁸ <a href="https://d197for5662m48.cloudfront.net/documents/publicationstatus/203973/preprint_pdf/415aa992aaf000b7940c9e06cc863bbf.pdf">https://d197for5662m48.cloudfront.net/documents/publicationstatus/203973/preprint_pdf/415aa992aaf000b7940c9e06cc863bbf.pdf</a></p><p>⁹ <a href="https://www.nextplatform.com/2023/05/02/the-skills-gap-for-fortran-looms-large-in-hpc/">https://www.nextplatform.com/2023/05/02/the-skills-gap-for-fortran-looms-large-in-hpc/</a></p><p>¹⁰ <a href="https://fortran-lang.discourse.group/t/frustrated-by-how-hrs-treat-fortran/5249">https://fortran-lang.discourse.group/t/frustrated-by-how-hrs-treat-fortran/5249</a></p><p>¹¹ <a href="https://www.reddit.com/r/AskProgramming/comments/1dwdpq3/language_for_scientific_computing/">https://www.reddit.com/r/AskProgramming/comments/1dwdpq3/language_for_scientific_computing/</a></p><p>¹² <a href="https://fortran-lang.discourse.group/t/the-counter-intuitive-rise-of-python-in-scientific-computing/469?page=10">https://fortran-lang.discourse.group/t/the-counter-intuitive-rise-of-python-in-scientific-computing/469?page=10</a></p><p>¹³ <a href="https://gmd.copernicus.org/preprints/gmd-2024-79/gmd-2024-79.pdf">https://gmd.copernicus.org/preprints/gmd-2024-79/gmd-2024-79.pdf</a></p><p>¹⁴ <a href="https://arxiv.org/html/2412.19770v1">https://arxiv.org/html/2412.19770v1</a></p><p>¹⁵ <a href="https://www.codee.com/optimizing-weather-research-forecasting-model-wrf-openmp-offload-codee/">https://www.codee.com/optimizing-weather-research-forecasting-model-wrf-openmp-offload-codee/</a></p><p>¹⁶ <a href="https://news.ycombinator.com/item?id=38891232">https://news.ycombinator.com/item?id=38891232</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=66c985bae4a2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Charting the Skies: The GRIB File Format]]></title>
            <link>https://medium.com/@julius.uy/charting-the-skies-the-grib-file-format-1ac4ea8a2cc2?source=rss-896373bd920f------2</link>
            <guid isPermaLink="false">https://medium.com/p/1ac4ea8a2cc2</guid>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[computer-science]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Julius Uy]]></dc:creator>
            <pubDate>Tue, 29 Jul 2025 06:12:15 GMT</pubDate>
            <atom:updated>2025-07-29T06:12:15.961Z</atom:updated>
            <content:encoded><![CDATA[<h3>Introduction</h3><p>Every time your phone notifies you that rain is expected in 12 minutes, it taps into a vast system of satellites, weather stations, radars, and numerical prediction models. These systems continuously generate and process petabytes of data, captured in dozens of unique file formats. But while the meteorologists analyze storm paths, weather conditions, and so forth, software engineers are the ones handling the infrastructure, storage, and parsing logic that makes all of this information usable at scale.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/420/0*0FG4YZRtIPYoXuCG" /><figcaption>Pew Research Center (<a href="https://www.pewresearch.org/journalism/2019/03/26/the-importance-of-local-news-topics-often-does-not-align-with-how-easily-the-public-can-find-information-about-them/">source</a>)</figcaption></figure><p>A 2019 study by the Pew Research Center found that, by a wide margin, weather updates are the most important type of current event information people rely on in their daily lives. One little-known fact is that weather forecasts aren’t just for deciding whether to bring an umbrella. They’re also essential for aviation. Even if conditions seem clear at ground level, the weather can be vastly different at cruising altitude — around 30,000 feet. Meteorological data helps determine whether a flight route is safe, when it’s best to fly, and whether flights should proceed at all.</p><p>Forecasters provide critical information such as headwinds and tailwinds, temperature shifts, ice accumulation (which can impair aircraft performance), global and atmospheric pressure variations, visibility, and localized advisories for pilots. These insights help aviation authorities and airlines make decisions about takeoffs, landings, and diversions.</p><p>Weather forecasting relies on a diverse network of instruments that collect data from the ground, air, and space. Ground-based tools like thermometers, barometers, anemometers, and rain gauges measure temperature, pressure, wind, and precipitation at fixed locations. Radiosondes — balloon-borne sensors — capture vertical profiles of the atmosphere, including temperature, humidity, and wind up to the stratosphere. Satellite instruments, both geostationary and polar-orbiting, provide continuous global coverage, collecting data on cloud cover, sea surface temperatures, and atmospheric moisture. Meanwhile, radar systems detect precipitation intensity and movement, which is crucial for short-term forecasting.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*p3kBCerQ-qJJy4NdIH3uog.png" /><figcaption>Your horoscope forecast… Cloudy with a chance of Comedy</figcaption></figure><h3>Specialized File Formats</h3><p>Weather and climate data isn’t just big data — it’s high-dimensional, real-time, geospatial, and heavily standardized.</p><ul><li><strong>Multi-dimensional</strong>: Data is not just time-series but also spatial (latitude, longitude, height).</li><li><strong>Frequent Updates</strong>: Forecasts update every half an hour to 6 hours; observations flow in every minute.</li><li><strong>Compression &amp; Transmission</strong>: Large models are shared globally via constrained channels.</li><li><strong>Metadata-Rich</strong>: Datasets must embed unit systems, grids, coordinate systems, and source provenance.</li><li><strong>Standardization</strong>: Agencies across the globe must interoperate reliably.</li></ul><p>This has led to the creation of specialized binary formats, cloud-native formats, and self-describing containers.</p><h3>The GRIB File Format</h3><p><strong>GRIB</strong> stands for <strong>GRIdded Binary</strong>.<strong> </strong>It’s a compact, efficient binary format for storing and exchanging large meteorological datasets like:</p><ul><li>Temperature</li><li>Wind</li><li>Humidity</li><li>Pressure</li><li>Precipitation</li><li>Weather forecasts and simulations</li></ul><p>Meteorological models produce huge volumes of data. GRIB was created to compress large datasets efficiently and allow fast access to specific variables, locations, and times. Moreover, because it is standardized by the WMO (World Meteorological Organization), it makes it possible to facilitate sharing between weather agencies worldwide.</p><p>A GRIB file is a sequence of messages. Each <strong>message</strong> contains <strong>one variable</strong> (e.g. 2-m temperature) at <strong>one forecast time</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*KKkhOUpsIC-wbSqS.png" /><figcaption>A GRIB file and its constituent elements (<a href="https://herbie.readthedocs.io/en/latest/user_guide/background/grib2.html">source</a>)</figcaption></figure><p>GRIB files are colossal. As a result, GRIB files are to be encoded in binary to allow for various compression algorithms to be utilised.</p><p>GRIB supports several templates, with the most common being <strong>Template 0 (Simple Packing)</strong>, <strong>Template 2 (Complex Packing)</strong>, or <strong>Template 40 (JPEG2000)</strong>. The choice depends on the data’s characteristics and desired balance between compression ratio and accuracy.</p><ul><li><strong>Simple Packing</strong>: Used for most scalar fields like temperature.</li><li><strong>Complex Packing</strong>: For fields with varying spatial variability (e.g., precipitation).</li><li><strong>JPEG2000</strong>: Lossy compression for fields where some data loss is acceptable.</li></ul><h3>Example GRIB File</h3><p>I am using a sample GRIB file for 28 July 2025, which can be downloaded from this link (the file is around 500 MB):</p><pre>wget https://nomads.ncep.noaa.gov/pub/data/nccf/com/gfs/prod/gfs.20250728/00/atmos/gfs.t00z.pgrb2.0p25.f003</pre><p>There are various tools to view a GRIB file. In this example, I use <a href="https://www.giss.nasa.gov/tools/panoply/">Panoply</a>.</p><p>Here’s what it looks like:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VqIOWfZmcd2a_U3h6Xs-Aw.png" /></figure><p>Because GRIB files are binary, the content is not humanly readable unless converted to some form, such as NetCDF.¹ The above example shows just that.</p><p>Below, I selected Temperature Surface and out comes a graph of how the world temperature looks like on that day:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*d0Q0tZLGWqCYR3vHUHTonA.png" /></figure><p>Behind the scenes, the data is represented in CSV format via Kelvin (To get the Celsius equivalent, subtract 273.15 from the Kelvin value).</p><pre>Latitude,Longitude,DateTime,Temperature_K,Temperature_C<br>40.7128,-74.0060,2025–07–28 11:23:00,298.15,25.00</pre><p>And here’s the actual 2D array equivalent for the entire world:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xFzEVJdPPwnVjwABV5djIw.png" /></figure><p>Tools such as Panoply allow one to look into various elements of a GRIB file and compare information across various days, such as year-on-year comparison on weather conditions, trend analyses, anomaly detection, and so forth.</p><h3>GRIB Subregions</h3><p>Earlier, I noted that a GRIB file contains a beginning byte for each message. This is important as it allows byte-range queries when a GRIB-index file is available. In the same way, one could query a timestamp of an MP4 to play (for example, dragging the timestamp of a YouTube video into somewhere in the middle allows you to skip watching portions of the video).</p><p>A byte-range query is therefore possible in the following form:</p><pre>curl -o outFile.grib2 --range 58224018-58456472 https://some-website.org/aaa.yyy.zzz.grib2</pre><p>In that manner, the size of the downloaded GRIB file is significantly diminished, allowing for faster processing.</p><h3>Conclusion</h3><p>The formats that weather data lives in are more than just storage containers. They shape how we process, analyze, and act on the forces of nature. As software engineers step more into data science, climate tech, and geospatial tooling, understanding formats like GRIB becomes a critical engineering skill.</p><p>Before the model, before the math — there was the format.</p><p>¹ I’ll discuss NetCDF in a future article (unless I get hit by a truck, in which case I hope I don’t reincarnate as a cockroach)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1ac4ea8a2cc2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How We Reduced the WINK+ Android App to Less Than 10MB]]></title>
            <link>https://proandroiddev.com/how-smrt-reduced-the-wink-android-app-to-less-than-10mb-00318f3e7d51?source=rss-896373bd920f------2</link>
            <guid isPermaLink="false">https://medium.com/p/00318f3e7d51</guid>
            <category><![CDATA[apk-size]]></category>
            <category><![CDATA[androiddev]]></category>
            <category><![CDATA[android-app-development]]></category>
            <category><![CDATA[android]]></category>
            <category><![CDATA[app-size]]></category>
            <dc:creator><![CDATA[Julius Uy]]></dc:creator>
            <pubDate>Sun, 29 Dec 2024 17:27:10 GMT</pubDate>
            <atom:updated>2025-01-06T12:07:42.759Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*B2d-1mKrGg9XPvXX" /></figure><p>One of the major challenges developers face in Android app development is optimizing app size without sacrificing functionality. Our team consists of only one Android Engineer working on the <a href="https://play.google.com/store/apps/details?id=smrt.media.wink&amp;hl=en_SG">WINK+ app</a>. Despite this resource constraint, we set an ambitious goal: to reduce our APK size to under 10 MB. A best-in-class product should look as good as it can be on the outside and the inside. Working with tremendous constraints forces us to think creatively to solve this. This is especially challenging when competing apps in the same category tend to be over 20 MB, and WINK+ offers much more complex features than them. One of the key bottlenecks we had was that reducing app size for the sake of reducing offers little to no value. Especially since we do not operate with abundant resources, we have to be very careful how we allocate them.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*1pEPl9Ss5VuQaVp_.png" /></figure><h3>1. Understand Your App’s Components</h3><p>The first step in any app size optimization process is to gain a deep understanding of the components that contribute to the app size. For WINK+, this included resources like images, libraries, and redundant code. Tools like Android Studio’s APK Analyzer help break down the app’s APK file and identify the largest contributors to its size. Unsurprisingly, the main culprits were large image files and unused third-party libraries.</p><h3>2. Remove Unused Resources and Libraries</h3><p>The most immediate win came from eliminating unused resources and libraries. Many apps include libraries that aren’t being fully utilized, either due to leftover features from earlier versions or third-party libraries that do more than required. We used Proguard, a tool built into Android Studio, to shrink and optimize the code. Proguard’s code shrinker helped remove unused classes, methods, and fields, reducing the overall size. Additionally, we conducted an audit to identify libraries that were too bulky for our use case.</p><p>One example is the QR Code Scanner, which we use for collecting WINK+ points all around Singapore. This is operated by MLKit under the hood. However, the MLKit library behind this takes up a healthy 2.4 MB of file size. However, because we only offer our app on Google Play and not on other third-party stores (which means all phones theoretically should have Google Play Services installed), we can <a href="https://developers.google.com/ml-kit/vision/barcode-scanning/android">unbundle</a> the library and rely on Google Play Services. This effectively shaved 2 MB of size for us.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/225/0*owLBfNdQERpx4oRE" /><figcaption>WINK+ users can scan QR codes all around Singapore to earn points in exchange for rewards</figcaption></figure><h3>3. Removing Unnecessary Fonts to Reduce App Size</h3><p>Fonts are another often overlooked aspect of app optimization. While they may seem small individually, fonts can add up quickly, especially when multiple font families or weights are included in the app. For the WINK+ app, we noticed that several unused font families and weights were being bundled into the final APK, contributing to unnecessary bloat. Removing these unused fonts helped us make a significant reduction in the app size without sacrificing the design or user experience.</p><p>We made sure to <strong>only include the essential fonts and weights</strong> that were required. Android allows you to define specific font weights and styles in XML using the <strong>font resource</strong> method, meaning that we didn’t need to bundle an entire font family with multiple weights and styles. By reducing the number of font variants and only keeping the ones actively used, we reduced the font asset size considerably.</p><p>Also, to avoid the need for downloading font assets from external servers (a process that can slow down app launch times), we embedded the fonts directly within the app, ensuring that the necessary font files were available offline. By carefully managing the fonts and removing the unnecessary ones, we were able to significantly reduce the size of the APK, and this optimization played a key role in keeping the app size under 8 MB.</p><h3>4. Opt for VectorDrawables Over Bitmap Images</h3><p>One major change that contributed to the reduced app size was switching from bitmap images to vector drawables. Bitmap images, especially those used for icons and other scalable elements, can add significant size to an app. Vector images, on the other hand, are scalable and typically much smaller in file size. By generally replacing PNGs with SVGs and other vector formats, we were able to dramatically reduce the amount of storage space dedicated to image assets.</p><p>However, one of the issues we had was that a lot of the PNG files never was on Figma, as we inherited the app from elsewhere, and working files were already lost. As a result, we relied on online resources to help convert PNG to SVG. Despite Adobe offering a <a href="https://www.adobe.com/express/feature/image/convert/png-to-svg">free service</a>, We found <a href="https://www.freeconvert.com/png-to-svg">this</a> to be the most effective. (Adobe ends up keeping a lot of strange artifacts)</p><p>We also took advantage of <a href="https://github.com/alexjlockwood/avocado">Avocado</a>, which allows us to further optimize the VectorDrawables by removing unnecessary points between lines, for example. However, some of the assets we migrated from PNGs had jagged edges, primarily because that’s generally how rasterized images end up being. We felt it was too much effort to clean that up, for a measly… few hundred bytes of savings, so we left them there.</p><h3>5. Optimize Dependencies and Modularize the Code</h3><p>Next, we took a close look at the third-party dependencies we were using. Many of these dependencies were large, adding unnecessary bloat to the app. One example was the Facebook Libraries.</p><p>We initially used this particular library, which loads up everything Facebook has:</p><pre>// Facebook Android SDK (everything)<br>implementation &#39;com.facebook.android:facebook-android-sdk:latest.release&#39;</pre><p>However, since we do not use everything Facebook have to offer, we opted for a lesser variant, thus saving us a tremendous amount of space as well.</p><pre>// Facebook Core only (Analytics)<br>implementation &#39;com.facebook.android:facebook-core:latest.release&#39;<br>    <br>// Facebook Marketing only<br>implementation &#39;com.facebook.android:facebook-marketing:latest.release&#39;</pre><p>By making sure that we only use dependencies that were strictly needed, we significantly reduced the app’s size. Techniques like “gradle dependency management” allowed us to strip down these libraries to their essentials, further optimizing the app’s build configuration.</p><h3>6. Compress Assets Without Losing Quality</h3><p>Asset compression was another vital step in reducing the app size. Tools like <a href="https://imageoptim.com/mac">ImageOptim</a> allowed us to reduce a significant amount of image file size by repurposing images with better compression.</p><p>Android Studio of course allows one to automatically convert images to WebP, whenever ideal:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/830/0*Had4y8K4aVzKlMRI.png" /></figure><h3>7. Optimizing Lottie Files for Faster Load Times and Smaller Size</h3><p>On WINK+, we use Lottie animations to enhance the user experience. Lottie files, which are JSON-based animations, have become a popular choice for app developers due to their lightweight nature and ability to scale without losing quality. They allow for smooth, high-quality animations that are much smaller in size compared to traditional video or GIF animations. However, even Lottie files can contribute to an app’s size if not optimized properly, especially if they include unnecessary layers or are too complex.</p><p>To reduce the overall size of Lottie animations used in WINK+, we employed several optimization strategies. First, we ensured that the animations themselves were as simple as possible. Lottie animations can sometimes include superfluous data, like extraneous keyframes or unnecessary assets embedded in the JSON file. Using the <strong>LottieFiles</strong> web-based asset optimizer, we were able to trim down these animations and remove unnecessary properties and redundant elements, making our Lottie files smaller and faster to load. We were, surprisingly, able to reduce around 1MB of file size just with this.</p><p><a href="https://lottiefiles.com/tools/lottie-json-to-optimized-lottie-json">Lottie JSON to Optimized Lottie JSON Converter - LottieFiles</a></p><h3>8. Test, Iterate, and Monitor</h3><p>Once the app was optimized and its size reduced to below 10MB, we conducted extensive testing. It was crucial to ensure that the app’s performance, features, and overall user experience were not affected. We continuously monitored app performance on different devices. In our case, we use <a href="https://newrelic.com/">NewRelic</a> for APM.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*56y295pME7SKUVwR" /><figcaption>Yay! 7.85MB!</figcaption></figure><h3>Conclusion</h3><p>Reducing app size can significantly boost download rates, as users often prefer smaller apps due to concerns about data usage, download times, and device storage. Research indicates that for every 6 MB increase in an Android app’s size, the install conversion rate decreases by 1%.</p><p>This trend is even more pronounced in emerging markets, where data costs and limited storage are major considerations. In countries like India and Brazil, a 10 MB reduction in app size can lead to a 2.5% increase in install conversion rates.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*iHZTpO9sT0Bt3vGA.png" /><figcaption>Install conversion rate increase per 10MB decrease in APK size by select market (<a href="https://medium.com/googleplaydev/shrinking-apks-growing-installs-5d3fcba23ce2">source</a>)</figcaption></figure><p>In conclusion, reducing the size of the WINK+ app to less than 10MB while maintaining a wider range of feature sets compared to similar apps was a challenging but rewarding process. By leveraging strategies such as optimizing resources, modularizing the code, using App Bundles, and focusing on efficient compression techniques, we were able to deliver a lean, fast, and efficient app. As app sizes continue to grow across the Android ecosystem, this experience has reinforced the importance of size optimization — not just for user convenience but also for improving app performance and engagement.</p><p>Happy coding!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=00318f3e7d51" width="1" height="1" alt=""><hr><p><a href="https://proandroiddev.com/how-smrt-reduced-the-wink-android-app-to-less-than-10mb-00318f3e7d51">How We Reduced the WINK+ Android App to Less Than 10MB</a> was originally published in <a href="https://proandroiddev.com/">ProAndroidDev</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Deliver High-Quality Software Products with the Agile Manifesto]]></title>
            <link>https://medium.com/@julius.uy/how-to-deliver-high-quality-software-products-with-the-agile-manifesto-4a20a72dd755?source=rss-896373bd920f------2</link>
            <guid isPermaLink="false">https://medium.com/p/4a20a72dd755</guid>
            <category><![CDATA[agile-methodology]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[project-management]]></category>
            <dc:creator><![CDATA[Julius Uy]]></dc:creator>
            <pubDate>Wed, 01 May 2024 12:15:28 GMT</pubDate>
            <atom:updated>2024-05-01T12:25:10.330Z</atom:updated>
            <content:encoded><![CDATA[<p>If one were to ask a product team to build from scratch…</p><ol><li><strong>A headless CMS with A/B testing,</strong></li><li><strong>A Natural Language Search written with the </strong><a href="https://en.wikipedia.org/wiki/Levenshtein_distance"><strong>Levensthein Distance Algorithm</strong></a><strong> with contextual prompt replacements baked in,</strong></li><li><strong>Migrating our views to use SwiftUI and Jetpack Compose </strong>and</li><li><strong>Build the product with an APM, Event Analytics, Digital Experience Analytics, infrastructure auto-scaling, and redundancy with a CDN supporting AVIF and JXL image delivery…</strong></li></ol><p>Do you think it can be built from scratch in <strong>five months</strong> with <strong>just four engineers</strong>? I seriously didn’t think it was possible until I was involved in a project that actually executed this from end to end, with those features baked in and more.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/0*tkSdXQI0SwfjEtuj.gif" /><figcaption>Me, living in constant despair over those five months of utter terror</figcaption></figure><p>One of the essential things to note is that in the Software Industry, engineers are often inundated by <strong>HOW </strong>problems but have too little a time to think through the <strong>WHY </strong>problems. This is why many engineers know <strong>HOW</strong> to solve the same type of problem devoid of a context but few have the depth to solve the <strong>WHY</strong> problem while addressing multiple meta problems branching off from it.¹</p><p>For example, when one looks at various tech events over the last few years, one wouldn’t find any shortage of materials on <strong>HOW </strong>to build with Gen AI, <strong>HOW </strong>to work on Vector Databases, <strong>HOW </strong>to build on blockchain, <strong>HOW </strong>to do this and that, and so on. Yet, the consideration of <strong>WHY </strong>is appallingly lacking. As such, this reflects the likely reason, among many others, why short-sightedness in tech has never been as evdient.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/333/0*-Wk-v7XIxTSm0NTp" /><figcaption>Me, back when I realized WHY I was a circus monkey</figcaption></figure><p>In this post, I’ll share some of the things we did to avoid these pitfalls. In fact, they are quite straightforward.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/602/0*pNkBzRJztoJ6vtnp" /><figcaption>Wrong. The most precious thing we all have is a 27-inch iPhone Pro Max Ultra</figcaption></figure><h3>Agile Manifesto vs. Agile Methodology</h3><p>When I was asked if we’re using Agile, I of course said yes. We are using agile software development. But I need to nuance it because when people talk about agile software development, they generally think that it’s the <strong>agile methodology. </strong>We do not use the agile methodology. Instead, we use the <strong>agile manifesto.</strong></p><p>While they may look eerily similar, the reality is that they are incredibly different. The <a href="https://agilemanifesto.org/">Agile Manifesto</a> is a way of thinking. The <a href="https://www.atlassian.com/agile">Agile Methodology</a> is a way of doing. The former is focused on the WHY. The latter, the HOW. Here’s where the problem comes, and Friedrich Nietzsche said it quite well: “If you know the why, you can live any how.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/850/0*mg8xdqXoMsWMCP4L.jpg" /><figcaption>This episode is brought to you by AZ Quotes and the German Super Mario. Danke</figcaption></figure><p>The Agile Manifesto in summary is essentially this: adopt the process to the outcome. The Agile Methodology on the other hand is this: adopt the outcome to the process. One doesn’t need too long to identify the colossal shortcomings of the latter. If it isn’t evident yet, it is this: a methodology is only as good as the context it is applied to.</p><blockquote>The AGILE MANIFESTO in summary is essentially this: <strong>adopt the PROCESS to the OUTCOME</strong>.</blockquote><blockquote>The AGILE METHODOLOGY on the other hand, is this: adopt the OUTCOME to the PROCESS.</blockquote><p>While we were working on this project, we didn’t do scrum nor kanban. We didn’t do proper PRDs. We didn’t document user stories and test cases. We didn’t use CI/CDs. No unit tests. No Git Flows. No Kubernetes. We test in production. IN PRODUCTION. We cut all that’s unnecessary and focused on the most important. We listened to customers, we listened to internal stakeholders, and our team talked a lot. As in, A LOT. Our standups can go as long as 90 minutes even though the proponents of the Agile Methodology (not the manifesto) demonize long standups.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/0*ZVBD2SVL8dnrICkX.gif" /><figcaption>A rare footage of Agile Coaches reading this for the first time</figcaption></figure><p>Why do we spend so much time talking in a standup? because a standup can evolve into an on-the-spot brainstorming and problem-solving. Because we have such a small team, the software engineers, the designer, and the product manager can directly discuss various constraints to generate the best outcome we could craft at that point. Because we talk a lot, we have a lot of shared context such that for the most part, every one of us knows why we are doing what we are doing, and in great detail. One can, for example, tap the shoulder of a random engineer in our team and he can explain in detail why we decided to do one thing the way we did it, and it’s not simply engineering. Whatever we do connects all the way to the Northstar KPIs.</p><p>Speaking of which, we don’t have an OKR system, despite John Doerr and the proponents of the OKR system championing it. We didn’t have the system in place because we didn’t need it.² We are agile. We talk a lot. We know the Northstar and we bridge the gap. This by the way is not possible without the agile manifesto. (or at least the ideas espoused by it) Which brings me to a very important point: Work From Home.</p><h3>No Work From Home</h3><p>One of the key articles of the Agile Manifesto is that it favors “Individuals and interactions over processes and tools.” It is no surprise that Steve Jobs for example, famously placed the restroom in the middle of Apple’s offices as well as cafeterias. This is to promote spontaneous interactions among employees henceforth generating rapport and consequently, creative ideas.</p><p>Work From Home takes this away totally. I need not repeat the research on <a href="https://en.wikipedia.org/wiki/Allen_curve">Allen Curve</a> and the consequences of it surrounding the lamentable state of Silicon Valley and its innovative spirit since the pandemic.</p><p>In our team, we don’t practice Work From Home. Of course, there are cases when people need to work from home. Those are generally reasonable. Some people work late nights and weekends. If they need some flexibility, we understand it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/220/0*_eg_7BwnkSQljhrQ.gif" /><figcaption>WFH is dead. Cry me a river, Bob</figcaption></figure><h3>Treating Everyone as Responsible Adults Until Proven Otherwise</h3><p>Speaking of flexibility, we also, therefore, treat everyone as adults. It is interesting to note that unproductive managers hide beneath the skirt of “Corporate Policies” and “HR Handbooks.” When the North Star is clear, everything else should contribute towards that direction. One should not sacrifice the success of the business in the name of corporate processes. Hence, when flexibility is essential for humans to be productive, then flexibility is offered. Why create a win-lose scenario when a win-win is possible between the employee and the business?</p><p>It is interesting to note why most managers fall under the trap of corporate policies and how they hide under it. It is primarily because these managers fail to think through the lens of First Principles.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/694/0*H55sHtWBkzqtEv9v.jpg" /><figcaption>This segment is brought to you by AZ Quotes and the Archbishop of Tesla</figcaption></figure><h3>Think with First Principles. Always</h3><p>First Principles thinking is simply this: When you are asked to do something, keep asking why until you cannot find further answer beneath the why. That, therefore, is the first principle. Understanding the first principles before one performs his job helps him identify the boundary conditions by which he can operate. This also means that because the first principles is the foundation by which all other decisions are made, he who knows them has the compass to navigate.</p><p>Most managers don’t know this. They don’t think like this, much less individual contributors. Which therefore means that they are then buffeted by whatever seems hot and trending in the marketplace chatter.</p><p>For the team therefore, despite the density of best practices out there, we had to ruthlessly prioritize what we could and could not do in the time and resource constraints we were given. Wouldn’t it be great if we had a dev, test, staging, demo, and production environment to work with? What about a proper CI/CD? Automated tests anyone? I hear music to my ears. Truth is, these are great. But in the size and scale that we’re in, putting these together slows us down. While the “Best companies out there” with the “best engineering culture” have many of these in place, in the end, when one is forced to do first principles thinking, are these going to improve the top line or the bottom line? How? If the answer isn’t clear, perhaps the need isn’t there, yet.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/0*Wq5mMkc1bU5sB5Ww.gif" /><figcaption>I have no idea why I included this GIF. I guess that’s… FIRST PRINCIPLES THINKING!</figcaption></figure><h3>What About Software Engineering Best Practices?</h3><p>It is interesting to note that throughout this entire article, I hardly wrote anything in detail about us doing anything fancy in our software development processes. This is primarily because we’re doing just about the very same thing every other average Joe is doing. We didn’t do anything special that you can’t read about on the internet or see your teammates do. In terms of software design, there’s nothing much to learn from me that has not already been said by others. It is, however, the agility to deliver quality output in a highly constrained environment that one can go beyond and think creatively.</p><p>Oh yeah, just in case you want to see what we built from scratch, here it is:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*c2wLi0ynC3RzyJPT" /><figcaption>Every five-star review you drop prevents me from being disowned by my family.</figcaption></figure><p>Android: <a href="https://play.google.com/store/apps/details?id=smrt.media.wink">https://play.google.com/store/apps/details?id=smrt.media.wink</a></p><p>iOS: <a href="https://apps.apple.com/sg/app/wink-singapore-mrt-metro-map/id1071049933">https://apps.apple.com/sg/app/wink-singapore-mrt-metro-map/id1071049933</a></p><p>___</p><p>¹ If there are any doubts, just look at the algorithm-type interviews still replete today.</p><p>² We didn’t need it, yet. We may need it when the team grows much bigger.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4a20a72dd755" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Tale of Two Tech Zeitgeists]]></title>
            <link>https://medium.com/@julius.uy/a-tale-of-two-teczeitgeists-daad049ac89f?source=rss-896373bd920f------2</link>
            <guid isPermaLink="false">https://medium.com/p/daad049ac89f</guid>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[tech]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[business]]></category>
            <category><![CDATA[startup]]></category>
            <dc:creator><![CDATA[Julius Uy]]></dc:creator>
            <pubDate>Sun, 24 Dec 2023 09:13:49 GMT</pubDate>
            <atom:updated>2023-12-24T09:31:46.052Z</atom:updated>
            <content:encoded><![CDATA[<p>I have spent the last two decades as a Software Engineer working on various products. I’ve worked on Application Tracking Systems in an MNC, Revenue Management for Airlines, VOIP, XMPP, and Social Media in a startup, Video Streaming, Construction, Education, and now, Transportation. Every single one of these companies presents certain challenges. Yet what was interesting between 2004 and 2023 is the striking difference between the principles governing the continued existence of products (and in many ways, tech companies) then and now.</p><p>The year was 2011 when I was working on an app called Hoiio. It’s a voice call and texting platform now more popular in services such as WhatsApp and Viber.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/590/0*P0Y4nwkrYkGLgYDC.png" /><figcaption>Hoiio Suite of Products, circa 2010–2012 (<a href="https://www.techinasia.com/hoiio-working-on-its-own-version-of-facebook-connect">source</a>)</figcaption></figure><p>Back in those days, WhatsApp didn’t offer calls and Viber didn’t offer chat. Both were free, which also meant they were unsustainable as a business. It was unclear to us at Hoiio how they wouldn’t end up going under. On the other hand, the service we offer was what we thought back then, a sweet spot between IDD calls that tend to cost an arm and a leg, and VOIP calls that cost less than half the price. This allows us to offer our customers IDD calls at a discount while also taking a service fee for every minute of call made.</p><p>On paper, the business model sounded good and was ripe for acquisition by a telco. Yet somehow, we ended up reaching a point where the cost structure didn’t make sense and we had to pivot. Yet surprisingly, two years later, in 2014, Viber was acquired at 900 million USD and WhatsApp, 19 billion. Had we stuck it out, would we have ended up with a multi-million dollar exit? No one knows.</p><p>That time, we had a realization. We focused too much on profitability that we were smothered by growth. Had we focused on growth, things might have ended up differently.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/444/0*jc2AN3kbbEXaf-fH.gif" /><figcaption>Here’s to growing my biceps eight days a week baby</figcaption></figure><p>I later moved to a video streaming company which ended up being acquired for 200 million in 2013. At that time, the company focused solely on growth and zero profit. Yes, a 200 million acquisition for zero profit. It boasts millions of minutes viewed monthly and thousands of daily active users. That company ended up being the leader of the pack, profitable, and now boasting billions of minutes viewed in a month.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/705/0*ZiMDtI6bbt-vYBFi.png" /><figcaption>My God, they didn’t include my face in this infographic! Oh well, it’s a face only a mother would love. (<a href="https://www.techinasia.com/visual-story/viki-mafia-team-building-startups">source</a>)</figcaption></figure><p>Did we do it right? Were we successful? All external metrics might say so. We were now the incumbent in our niche and other players are not anywhere near. Yet, would we have ended up being acquired for 200 million with zero revenue in 2023? I’m not so sure.</p><p>Today, we now observe a vastly different story. As early as the first quarter of 2022, Y Combinator and various accelerators are drumming the beat that startups should focus on saving for rainy days. In those days, Michael Seibel and Jason Calacanis, if my recollection was correct, noted that startups should make sure they have at least 24 months of runway.</p><p>By the tail end of 2022 and for most of 2023, tech entered a layoff bloodbath not seen since the dotcom crash of 2001. Today, the narrative has changed. From venerating $250,000 TCs for fresh graduates and free massages, we have now moved our perspective back to what’s most important for business: profitability.</p><p>Many who have not experienced the dotcom crash nor the global financial crisis of 2008 lived in a Disneyland where free money and low-interest rates turned the tech industry into a highly spoiled employee’s market. Land grabs and growth at all costs have rendered the sprint to profitability one which founders continued to kick the ball down the road, wishing they could cash out before the first principles of capitalism emerge for the reckoning.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/0*2LJ1oHpHrUSQ-cfb.jpg" /><figcaption>Susan, I need a Hibiki 21 to wash away my sorrows</figcaption></figure><p>Today, we observe a world where cognitive flexibility and generalism trump specialism. What value is one’s deep understanding of data science where the top of the funnel has too many applicants than the job openings waiting to be filled? Moreover, how does one really justify the attribution of paying a Data Science staff to a revenue bump, let alone, a profitability bump? We never found a means to justify that when I was in Revenue Management… and to be clear, increasing revenue by 1% for airlines, in general, meant multiple millions in revenue.</p><p>This brings into question the entire decade of engineering odysseys at peacetime. How many new programming languages are good enough? Proponents of microservice architecture have demonized monoliths for nearly a decade, only to realize that monoliths for the most part served their purpose, and served them well. Agile proponents perpetually demonized waterfall, only to now realize that certain frameworks are applied best in contexts they’re best for. We over-engineered what was meant to be simple, and we oversimplified what was meant to be complex.</p><p>One would be surprised by the amount of time engineers have told me the importance of understanding multiple metas beneath abstractions because this is how good engineers should think, failing to realize of course that the reason why abstractions exist is to reduce the cognitive load. Minimizing cognitive load in understanding structures opens up more time for the engineer to be productive elsewhere, such as delivering features and stabilizing the product.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/498/0*J5ZlFd9kvBvDva1u.gif" /><figcaption>Mein gott! What have we done with our lives?</figcaption></figure><p>Engineers for the past half a decade have also been vastly romanticized by the need for titles: a chasing after the wind. The likes of Deloitte, JP Morgan, and Accenture have long known that these titles are useful to keep the juniors in a loop of revolving doors to manage their costs. Of course, young engineers can only chase after what’s readily available. Singapore, India, and Indonesia, among others, have been readily giving away Staff-level and Director-level roles like hotcakes. Only to of course later realize that engineers are over-levelled for the skills they have.</p><p>Moreover, one would be surprised at the amount of money engineers ask in Singapore, especially in 2023. It wouldn’t be far-fetched to see fresh graduates ask for six-digit salaries out of school. Of course, at times when growth at all costs and interest rates are low, land grab is the name of the game. At times when interest rates are high, survivability is the name of the game.</p><p>One must not be quick to blame the young engineers for being greedy. They live in a zeitgeist where everyone’s going around pushing salaries up, claiming gargantuan TCs without easy means to verify. Moreover, recruitment firms are likewise incentivized to rack up the numbers. The better the published numbers, the better the top and bottom lines end up.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/0*YUjoRqMWpdYUjD53" /><figcaption>(<a href="https://www.reddit.com/r/memes/comments/zjm7le/average_sam_bankman_fried_moment/">source</a>)</figcaption></figure><p>At the end of the day, what we observe is a perpetual tug of war between employees and employers, supply and demand, businesses and businesses. Such is the game of capitalism.</p><p>While one might quip that 2023 should serve as a lesson to those who lived it through, the same phenomenon is bound to happen a decade from now, or so. It is a cycle, one of which not a single professional is impervious.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/474/0*PLOzm0D39JaMuon6" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=daad049ac89f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What Makes a Good Employee?]]></title>
            <link>https://medium.com/@julius.uy/what-makes-a-good-employee-fc8397aa3730?source=rss-896373bd920f------2</link>
            <guid isPermaLink="false">https://medium.com/p/fc8397aa3730</guid>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[management]]></category>
            <dc:creator><![CDATA[Julius Uy]]></dc:creator>
            <pubDate>Sat, 19 Aug 2023 07:23:03 GMT</pubDate>
            <atom:updated>2023-08-19T07:23:03.329Z</atom:updated>
            <content:encoded><![CDATA[<p>Time and again, in every company I have taken some form of leadership role in, the axiom I’ve found to remain true (and expect to be so) is this: <strong>The business must win</strong>. Every action, every decision, be it sales, marketing, product, design, engineering, operations, and all the way back to sales has to have this axiom in mind: the business must win.</p><p>In many a place, especially in large organizations, there is a tremendous bias for inaction and unproductive bunch chasing their tails. This is far different from a startup. Startups don’t fail fast and fail often. In startups, failure means death. Startups don’t sing kumbayah believing that money grows on trees. When startup employees sit on their hands, it’s only a matter of time before it meets its demise.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*UiV56qjiFDwytpmk.jpg" /><figcaption>Study shows that sitting on hands long enough causes one’s business to end up the market leader</figcaption></figure><p>In large companies of course, especially when the engines of monetization have already been built by those who came before, there is a significant amount of room to nurse inaction and incompetence.</p><p>It is interesting that <a href="https://www.amazon.jobs/content/en/our-workplace/leadership-principles">Amazon’s leadership principles</a> <em>ownership </em>and <em>bias for action</em> are often found missing in many large organizations where people think that playing it safe is playing it well. In essence, indecision and deflection of ownership seem to be regarded as some form of corporate survival mechanism which, obviously, will last only for so long as the onion is left unpeeled.</p><p><a href="https://sg.finance.yahoo.com/news/snowflake-ceo-why-you-must-declare-war-on-your-competitors-183109540.html">Business, in its simplest form, is war</a>. Your competitors are out there waiting to take market share from you and run you to the ground. The office is not the place to play ping pong at 11 am and have 3 hour lunches. The office is a battlefield. In war, you don’t put officers in charge to sit on their hands because those who do have better use as meat shields than taking up space.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/498/0*soGX1thQCd1xLm1B.gif" /><figcaption>Somebody get Linda to unleash the Kraken</figcaption></figure><p>There is, of course, a sizable amount of people in companies large and small who refuse to make calls and kept waiting for the boss to tell them what to do. If the boss were to tell his engineers what to do all the time, there is obviously an option to replace them with someone who’s going to tell the boss what they’ll do and why. In a business replete with knowledge workers, there’s little room to hire yes men. It is a common axiom that a boss is to hire those better than them to fill the gaps. Hence, if the boss ends up better than the subordinate in the task he ought to be responsible for, what value then, does the subordinate contribute?</p><p>Elsewhere was a research on <a href="https://hbr.org/2023/02/what-is-psychological-safety">psychological safety</a> whereby it has been found to be one which significantly contributes to worker productivity. What the research failed to show, of course, is that this is only true when the worker has already enough self-confidence that he’s the right person for the job. A wrong person for the wrong job will find it excruciatingly hard to feel psychologically safe.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/0*2FX-VpGva2nrZmG1" /><figcaption>Managing Marketing as a Marketing Manager doesn’t make me feel safe</figcaption></figure><p>This of course means that the group of people you work with is a sports team and not a family. You cannot just anyhow disown a family member. Yet in a sports team, you can quickly replace a non-performer and a bad apple. Simply put, when the business culture trivializes the cuddling of bad behavior and poor performance, what is tolerated becomes normal and what becomes normal becomes the culture. So a leader needs to be firm but personable, decisive yet reasonable.</p><p>It is common knowledge now that delaying a decision to let go has never been a good decision. The boss often ends up wasting so much money and mental real estate managing toxicity and incompetence. To be clear, toxicity is an immediate reason to let go. Incompetence is usually not the person’s fault. It is usually the hiring manager’s fault.¹</p><p>In business, <strong>excellence is the expectation, not the exception</strong>. Yet many a time, one of course observes this in the opposite.</p><p>What makes a good employee? It is he who gets the first principles right and acts on them. The answer really, is quite simple. He only simply needs to ask, “Am I creating more value than what I am paid for?” The challenge of course is that most people can’t give a psychologically safe answer for this. Netflix was right in that companies do not need tons of people. Companies need very few people who are excellent in what they do. Yes, excellent people cost more than average people, yet they deliver value in greater multiple than the average joe. To run a good business is to run it meritocratically, not communistically. Therefore, it is pertinent that the right roles are filled by the right people. The business must win, and excellence is to be expected.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/886/0*91_I0bcpedz9J8KQ.gif" /><figcaption>An excellent Bollywood intermission to make sense of none</figcaption></figure><p>___</p><p>¹ of course caveated by all sorts of reasons such as inheriting a team member, reassignment of duties, etc.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fc8397aa3730" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Can Product Companies Thrive in the Singapore Tech Scene?]]></title>
            <link>https://medium.com/@julius.uy/can-product-companies-thrive-in-the-singapore-tech-scene-8fd7aa177435?source=rss-896373bd920f------2</link>
            <guid isPermaLink="false">https://medium.com/p/8fd7aa177435</guid>
            <category><![CDATA[business]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[entrepreneurship]]></category>
            <category><![CDATA[singapore]]></category>
            <category><![CDATA[startup]]></category>
            <dc:creator><![CDATA[Julius Uy]]></dc:creator>
            <pubDate>Sun, 12 Mar 2023 09:15:50 GMT</pubDate>
            <atom:updated>2023-03-12T09:15:50.366Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/843/0*PZRvWrHkAUfnpugy" /><figcaption>Me flexing my awesome Samsung Photography Skills</figcaption></figure><p>Back in the infancy days of Carousell, Shopback, Lazada, and Shopee (circa 2012 to 2015), and several others that went before them, like my previous startup <a href="https://www.techinasia.com/hoiio-working-on-its-own-version-of-facebook-connect">Hoiio</a>, the Singapore tech scene as compared to where it is today has seen some tremendous shift. In the most recent NUS survey, for example, we have seen that fresh grads in Computer Science <a href="https://www.moe.gov.sg/-/media/files/post-secondary/ges-2022/web-publication-nus-ges-2022.ashx">earn as much as $7,000</a>. Separately, we have also seen students hoping to be paid <a href="https://www.asiaone.com/singapore/10k-starting-pay-netizens-express-disbelief-nus-undergrads-salary-expectations">between $9,000 to $10,000</a> as soon as they’re out of school.</p><p>It’s fascinating how the trumpet call in the last several years has mostly been about “how much one can earn” rather than “how much impact one might have.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/0*zFOBD15XhUao6Ghc.gif" /><figcaption>I’m still going to build this shower in my flat. STOP JUDGING!</figcaption></figure><p><strong>Back in the infancy days of Carousell and Shopback, the pitch to software engineers was not about them having an amazing salary package, but rather how flat the hierarchy is. </strong>In essence, a Software Engineer can straight up talk to the CEO to bring up ideas. Likewise, back in those days, Singapore has not yet become a breeding ground for software best practice zealotry. For example, when I stepped into Singapore back in 2010, I was appalled at how Test Driven Development, Code Reviews, and Software Architecture were done (or the lack thereof). I came from a much bigger US tech firm with clearly defined processes and following enterprise software architecture design I used to think that Singapore was behind. Yet in hindsight, I realized that the scrappy nature of Singapore back then was what fueled Carousell and many others to where they are today.</p><p>Back in Hoiio, and even in a much bigger company like SingTel which I joined later, we didn’t do TDD, we didn’t have proper DevSecOps, zero APMs, no JIRA, no Slack, no PagerDuty, no marketing automation. But what we had was speed and cost efficacy. We can build really really fast at really really cheap. Sure, one can argue that the quality is not there. Yet the counterargument was that maybe you don’t need that high a quality!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/0*hmRE9pQNnCoJchY-.gif" /><figcaption>Nobody cares what you think if you’re not giving me the moolah!</figcaption></figure><p>Today, starting a product company in Singapore is often met with unfortunate desolace. Just look at the sheer number of product startups in Singapore and how poorly funded/staffed they are. For products that aim to impact a global audience, chances are, America and China would have had economies of scale and better-capitalized companies already playing in the scene. On the other hand, making a regional play is not any better. Singaporeans do not understand the Indonesian, Philippine, Thai, and Vietnamese markets well enough to monetize in the region. (Localization efforts notwithstanding) Moreover, the high-quality product expectations in Singapore do not directly translate to the needs of their regional counterparts. Hence, Singapore companies overspend on trivial matters.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/0*YLK2kO0rVwPY21v1.gif" /><figcaption>Who cares if we can print money like nobody’s watching??</figcaption></figure><p>This is not unique, however. Singapore has never been good at regional plays. Given the small TAM in Singapore (barely six million at most), Singapore companies have no choice but to go regional, and fast. And while one might think that B2B Product Companies might work, they will only go so far until their American and Chinese counterparts enter the market. (A saving grace of course is that Singapore is often too small a market to prioritize)</p><p>Singapore therefore, in general, seems to be losing its tiny hope to make a name for itself as the “Silicon Valley of Southeast Asia.” The incentives and market conditions work against tech entrepreneurs. Of course, the situation gets even worse in places like the Philippines, Myanmar, Cambodia, and Laos. At least Singapore can poach talents from these countries. On the other hand, Indonesia and Vietnam are, if things continue to move the way they do, likely going to take the crown off Singapore as the innovation hub of the region.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/220/0*Q5XuzgRo13pu7w_u.gif" /></figure><p>Singaporeans, being very western in their influence, have often complained about cultural mismatch with Chinese companies. Indonesians and Vietnamese on the other hand, have been brushing up very quickly in their command of English. Hence, it seems that it’s just a matter of time before Singaporean Software Engineers lose both their cultural and linguistic advantage.</p><p>Singaporean Software Engineers, therefore, are likely going to be left with a few options. One, they can enjoy the benefit of their citizenship and work in the US with the H1B1 visa. (and really make a difference working on core products) They can also consider moving to other roles, such as Project Management. Finally, they can continue working on tier 2/tier 3 projects in Big Tech. Of course, all these lead to brain drain.</p><p>It’s unclear what the future holds. But without a strong zeitgeist to put together smart software engineers with well-capitalized startups, Singapore is likely going to lose its crown as a tech innovation hub in the next decade or two. But then again, even Nostradamus can’t get his stock market predictions right, much less this armchair analyst.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/0*3IU8gH5041bGVRhE.gif" /><figcaption>Susan!!! The Crystal Ball says I need to buy a Maserati on Tuesday</figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8fd7aa177435" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[So You Want To Be An Engineering Leader?]]></title>
            <link>https://medium.com/big-on-development/so-you-want-to-be-an-engineering-leader-8845bef658a7?source=rss-896373bd920f------2</link>
            <guid isPermaLink="false">https://medium.com/p/8845bef658a7</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[engineering-mangement]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[careers]]></category>
            <dc:creator><![CDATA[Julius Uy]]></dc:creator>
            <pubDate>Sat, 31 Dec 2022 03:50:40 GMT</pubDate>
            <atom:updated>2023-01-01T08:01:09.157Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/640/0*Yw4SbpiAEsqaOYsM.gif" /></figure><p>For twenty years now, the Spider Man quote, “with great power comes great responsibility” has been used so much that it deserves no further veneration. Software engineers for the past ten years found themselves in a rather unusual spot. Not only are they paid a ton, the cost of failure in anything they do has very little bearing. When a surgeon screws up, life is at stake. When a software engineer screws up, all that’s needed is a hotfix. I say that because a Software Engineer earns around as much as a surgeon yet the former found himself cuddled with bean bags and free food and ping pong tables.</p><p>This phenomenon of course, which is supposed to be a breeding ground of gratitude, ended up being a breeding ground of greed (or as <a href="https://www.cnbc.com/2022/12/10/billionaire-charlie-munger-world-is-driven-by-envy-not-greed.html">Charlie Munger rightly pointed out recently… envy</a>.) Instead of celebrating their positions in this capitalist society, software engineers ended up, like any other normal human, fighting over dominion… and with that comes titles.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/0*QFY2mlmbhJV6yQMU" /><figcaption>Presenting, the Senior Chief Principal Vice President Director of Software Engineering</figcaption></figure><p>Back in those days, there are no such things as Staff Engineers, Principal Engineers, Distinguished Engineers, Fellows, and so forth. There are just Software Engineers… and we’re perfectly happy with it. I used to work in a company where most of us had eight to twelve years of experience and our titles are… you guessed it… <em>Software Engineer</em>. No Senior, no Staff.</p><p>Yet the world has changed. A new generation of engineers who haven’t been through the 2000–2001 dotcom crash and the 2008–2009 Global Financial Crisis created an industry of entitlements… a chasing after the wind, which then of course created in itself some of the rather toxic hiring processes and even work cultures. It turns out that adversity does not build character, it reveals it.</p><h3>Adversity does not build character, it reveals it</h3><p>Looking at career matrixes, I have yet to see one that attributes <em>maturity</em> to <em>seniority.</em> How <em>mature</em> must your Principal Engineer be? Does he roll on the ground screaming like a two-year-old when things don’t go his way? What about your Director of Engineering? Or your CTO? Does he have the backbone to be the bad guy? Can he make a decision not everyone will be happy with? Can he take criticisms, more specifically, those unfair ones… and move on?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/268/0*1Gio8hTcd5fBBVdN.gif" /><figcaption>Who’s angry?? I’m NOT ANGRY!!!!!!</figcaption></figure><p>It’s rather interesting that software engineer interviews have a lot to do with detecting technical skills (nothing wrong with that) but too little to do with detecting their <em>maturity. </em>Software Engineering, it turns out, is as much a technical endeavor as it is a team effort.¹</p><p>Looking at the inventory of Software Engineers I’ve worked with, from Junior all the way to VP/CTO levels, I’ve seen around 15–25% of them have problems with emotional maturity. While this can be sandbagged better for the individual contributors, those who reach the Middle Management and Senior Management levels end up becoming trouble.</p><p>The Software Engineer’s primary job is to help the business win. His secondary job is to write good code. An immature software engineer will flip those two around. For the business to win, there should be alignment between people and between teams. As much as the engineer hates meetings, that’s exactly what is needed to drive alignment. If meetings are expensive, try misalignment.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/440/0*Xd4P0nEkwrDHyQ8O.jpg" /><figcaption>Meetings or This?</figcaption></figure><p>Which leads me to another point.</p><h3>Fire Fast, Hire Slow</h3><p>If there’s one thing I learned in my years as an engineering leader, it’s that at the moment I lost hope on a direct report, there’s no point in not firing the person. I’ve always found that delaying a decision not only adds more suffering to the person, it also adds more suffering to the rest of the team. Firing is not fun, that’s why most people want to avoid it.</p><p>Over the course of my career, I’ve recommended several dozen to be let go. Those whose recommendations were denied end up causing more trouble. Some were let go later. Others demoted. Still others were left without the senior management lifting a finger, hence thus, remains a festering problem.² Is it fun? It’s never fun. What reasonable human feasts on the suffering of others?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/240/0*aqj3yRActP1M6ZPZ.gif" /><figcaption>He’s fired. Literally.</figcaption></figure><p>I’ve also found quite a bit of managers (some very experienced ones), who dread firing. If the engineering leader hasn’t built a backbone to stand up to criticism, he should not be an engineering leader. Tech Leads, Engineering Managers, Directors, VPs, and CTOs are magnets of criticism. The higher you go, the worse they are. The higher you go, the more people you will take livelihood away from… because unlike being a Software Engineer, you’re now sitting in between the business and the person and you have to often make difficult decisions. Sink the ship, or throw one overboard? <strong>The engineering leader’s main responsibility is toward the people that should stay, not the people that should go. </strong>Often, there are no easy decisions, and the night after you let someone go… the soup will taste like corpses.³</p><p>This, therefore, means that one should be very careful with hiring.</p><p>I’ve made a mistake myself whereby I felt strapped with the time to fill and, in wanting to appease a colleague, ended up hiring someone who wasn’t a good fit. I also learned along the way that of all the people we ended up hiring that I had doubts about, and there are dozens of them, only two performed to, or above my expectations. Every one of them I wished we never hired.</p><p>I learned that a colleague who fails to respect the overall opinion of the hiring committee either has to change attitude or change company. I did worry about this in the past and I may end up making the same mistake in the future… yet every time I worry that someone smart might harbor bitterness because he didn’t get what he wanted, I end up being instrumental to an outcome nobody wanted. In essence, I could have prevented a sour outcome and chose not to.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/498/0*VkNiDB5r6sjoeOst.gif" /></figure><h3>Leaders Lead. Coders Code. Learn the Difference</h3><p>Another common problem I’ve seen is that many engineering leaders can’t let go of coding. Every hour an engineering leader spends coding is an hour he could have spent leading. Leading requires vast and thorough strategic thinking, influencing, driving alignment, executing, and measuring.</p><p>If the engineering leader can’t let go of coding, one better believe a dysfunctional team is buried beneath the surface. To be clear, a dysfunctional team is not always a team of unhappy people. Rather, it’s a team of underutilized talents and suboptimally deployed resources.</p><p>So if the engineering leader can’t let go of coding, he should not be a leader. There’s no shame in handing over the mantle to someone who can do a much better job.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/639/0*9jwuswWiawN7o4k4.gif" /></figure><h3>Conclusion</h3><p>So you want to be an engineering leader? With great power comes great responsibility. Life doesn’t get easier the higher one goes. The stakes are higher, the problems are tougher, and often, there are no easy options. If the engineering leader can’t take unfair criticisms, can’t fire people, and can’t stop coding, he’s probably not ready.</p><p>Now that 2023 is coming, if you haven’t soiled your hands with blood as an engineering leader, be prepared. 2023 will be a tough year in tech. If you’ve been cuddled all these years and didn’t have to do the dirty job, prepare a garbage bin. You might need to <a href="https://nypost.com/2022/11/11/twitter-manager-vomits-after-elon-musk-orders-him-to-fire-staff/">puke on it a few times</a>. There’s a very thin line between charity and capitalism. Idealism and realism. The wish is that you can have both… yet many a time, you have to choose one at the expense of the other… that’s… not fun.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Rgk2QAuIaO_Pi9xo" /></figure><p>___</p><p>¹ For the most part. Unless of course you’re the sole engineer in your company, in which case, you can code in Wenyan and nobody cares.</p><p>² The truth, of course, is always somewhere in the middle. Which means my opinion does not fairly represent the aspects that I don’t see.</p><p>³ Pulled from Elie Wiesel’s Night. Go read it.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8845bef658a7" width="1" height="1" alt=""><hr><p><a href="https://medium.com/big-on-development/so-you-want-to-be-an-engineering-leader-8845bef658a7">So You Want To Be An Engineering Leader?</a> was originally published in <a href="https://medium.com/big-on-development">Big O(n) Development</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Rethinking Android Runtime with Profile-Guided Optimizations]]></title>
            <link>https://proandroiddev.com/rethinking-android-runtime-with-profile-guided-optimizations-a1179cde5746?source=rss-896373bd920f------2</link>
            <guid isPermaLink="false">https://medium.com/p/a1179cde5746</guid>
            <category><![CDATA[compilers]]></category>
            <category><![CDATA[androiddev]]></category>
            <category><![CDATA[android]]></category>
            <category><![CDATA[android-app-development]]></category>
            <category><![CDATA[optimization]]></category>
            <dc:creator><![CDATA[Julius Uy]]></dc:creator>
            <pubDate>Sun, 13 Nov 2022 06:24:28 GMT</pubDate>
            <atom:updated>2022-11-14T01:36:03.061Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*s0wXnkOBkOrNSVnU.jpg" /><figcaption>I should buy a boat</figcaption></figure><p>Recent conversations around Android Runtime Optimizations has to do with a technique known as Profile-Guided Optimization. (or <a href="https://www.ibm.com/docs/en/aix/7.2?topic=techniques-profile-directed-feedback">Profile-Directed Feedback</a>) The idea behind it is that the software engineer can consider various code paths a user would typically go through and pre-compile these paths instead of leaving it to the JIT compiler. This is certainly interesting but not without caveats. But before we dive into that, we should do a quick review of how Java works.</p><p>What makes Java interoperable across various OS and CPU architectures is because of the JVM. How Java compilers work is that they convert the Java source code into bytecode. The bytecode is then interpreted by the JVM and is converted to machine code depending on the OS and CPU architecture it runs on. Below is how the whole process looks like:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/549/0*OdvHOLCeEqB0-swH.png" /><figcaption>Source: <a href="https://www.scientecheasy.com/2021/03/what-is-jvm.html/">https://www.scientecheasy.com/2021/03/what-is-jvm.html/</a></figcaption></figure><p>An Android engineer might know by now that Android does not use JVM. Because Android runs on memory and space-constrained devices, Dalvik is used in its stead (and since Lollipop, ART)¹ Below is the modified Android compilation and code execution path:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/612/1*k0vofaj723hOZPpqM8IDpA.png" /><figcaption>Source: <a href="https://medium.com/programming-lite/android-core-jvm-dvm-art-jit-aot-855039a9a8fa">https://medium.com/programming-lite/android-core-jvm-dvm-art-jit-aot-855039a9a8fa</a></figcaption></figure><p>Unlike Dalvik which depended on JIT (Just in time Compilation), Android Runtime depended largely on AOT (Ahead of Time compilation). How JIT works is by having the Java bytecode converted into machine code at runtime. This allows Java to remain interoperable and also usable, especially in storage-constrained environments.</p><h3>AOT to the Rescue?</h3><p>One might ask, “if JIT compiles the code at runtime, wouldn’t it be slow?” The answer is yes. JIT sacrifices performance over storage. (Not entirely true. More on this later) Of course, a counterproposal has to be raised. At a time when storage is now so cheap, can we then optimize for performance over storage? Those who thought such would then flock to Ahead of Time Compilation (AOT). In this strategy, Java pre-compiles various parts of the code so that the JIT doesn’t have to take that responsibility. The drawback of course is that because AOT code has to be executed across various operating systems and CPU architectures, all of these pre-compiled machine code has to come together under the same java package, thereby bloating up the file size.</p><p>This is not a problem when the OS is not often updated. However, because Android and its OEMs regularly patch their OS, precompilation has to be done every time there’s an update. This leads to a significant amount of wait time for users who have just updated their phones.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/0*PS8ce3GqDto5yvf7" /><figcaption>This is why I can’t play Genshin Impact with a smile on my face</figcaption></figure><h3>Enter Crowdsourced Data</h3><p>Now let’s go back to JIT for a second. More modern implementations of JIT comes with an execution counter so that it knows what part of the code path to pre-compile, hence improving performance. This of course is based on how the individual user interacts with the application. Now, suppose the JIT has access to the cloud, such that it could crowdsource the code paths used by various users around the world. Could it, in theory, pre-optimize the execution for the user? The answer is yes. Now that the cloud is widely available, JIT can rely on crowdsourced data on the cloud to make apps run even faster. For Android, Google calls this Cloud Profiles.²</p><p>Cloud Profiles, however, has one problem. While they excel at identifying execution paths, they are only as good as the lifespan of the app version. When the app is updated, Cloud Profiles for the previous app version are rendered useless as shown below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-N5yu21w7-k8xujN3kjTVw.png" /><figcaption>Source: <a href="https://youtu.be/yJm5On5Gp4c?t=438">https://youtu.be/yJm5On5Gp4c?t=438</a></figcaption></figure><p>This is because Cloud Profiles won’t have data about how the new app version is executed. Hence, is unable to pre-optimize for early-access users. Of course, this will taper over time, but the problem remains for early adopters, and if the company does weekly deployments, such as Twitter and Netflix, the benefits of Cloud Profiles are diminished.</p><h3>Profile Guided Optimizations?</h3><p>Recently, Android introduced Profile Guided Optimizations through the use of <a href="https://developer.android.com/topic/performance/benchmarking/macrobenchmark-overview">Macrobenchmarking</a> and <a href="https://developer.android.com/topic/performance/baselineprofiles/overview">Baseline Profiles</a>. This is to solve the problem mentioned above. In essence, the idea is to find the right balance between JIT and AOT so that only the code paths the software engineer thinks users often go to are optimized. If the AOT codes come shipped with the app, then one can find the right balance between having just the right AAB file size and also having the app run as smoothly as it can be. But there’s one problem.</p><p>Consider Grab, which surfaces various home screens depending on which country you’re in:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3mEb4G5_8XCDCZzOne4wUg.jpeg" /><figcaption>Grab Homepage in Singapore, Vietnam, and Indonesia.</figcaption></figure><p>Here, Profile Guided Optimizations lose their appeal. Because now, shipping profiles tied to each country tends to be sub-optimal. Moreover, when the use case branches out into various features users might use in a super app, the value of Profile Guided Optimizations quickly diminishes.</p><p>In fact, this <a href="https://www.youtube.com/watch?v=gx8DVVFPkcQ">2019 talk</a> noted that AOT tends to be almost always suboptimal compared to JIT because it relies on highly specific code paths which may not always be true in actual usage.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TIIhBih8XM9iH9fEORGLAg.png" /></figure><p>Given enough time, JIT will always outperform AOT since it has a larger pool of data to use. Moreover, one does not have to predict the range of code paths users will go through prior to them actually going through them.</p><h3>What’s Next?</h3><p>In the end, all is not lost. Because most Android apps come shipped with Firebase and run on devices with Google Play Services, these two can coalesce and send common usage data to the cloud, as they already do with Firebase Analytics, Firebase Test Lab, and Android Vitals. A user’s typical usage patterns won’t vary across app versions, assuming that the means to do so remain intact, the data from these services can help perform automated profile-guided optimizations based on large volumes of data. Hence, further optimizations for app performance are likely to happen on the Google Play side of things where Cloud Profiles will attempt to pre-optimize areas of optimal usage that are personalized in nature. This of course is a storage hog.³ Yet now that storage is significantly cheaper than it used to be, incrementally personalizing this runtime optimization is certainly not far-fetched.</p><p>In conclusion, one is advised to be mindful of using <a href="https://developer.android.com/topic/performance/baselineprofiles/overview">Baseline Profiles</a> even when the recent Android Dev Summit and various online resources push for this. For Southeast Asia where super apps are quite common, this can come at a significant human capital expense with little to no benefit. If Google Maps increased searches <a href="https://android-developers.googleblog.com/2022/01/improving-app-performance-with-baseline.html">by only 2.4%</a>, and it’s one of the top of the funnel activities, improvement on essential conversion metrics, such as purchasing a product, is even much lower. So if it’s not worthwhile to spend hundreds or even thousands of dollars of man-hours for baseline profiles, I recommend holding back. Cloud Profiles will be that which would save the day, and you barely have to lift a finger.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/220/0*NbLIQsGDyvor_c9u.gif" /><figcaption>Susan!!! The remote is out of battery!</figcaption></figure><p>___</p><p>¹ Technically you can turn on ART on KitKat via Developer Options but it didn’t become the default option until Lollipop.</p><p>² Of course, the Java world has various alternatives such as JITaaS.</p><p>³ But Google just gave everyone a free 1 TB storage so there’s that.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a1179cde5746" width="1" height="1" alt=""><hr><p><a href="https://proandroiddev.com/rethinking-android-runtime-with-profile-guided-optimizations-a1179cde5746">Rethinking Android Runtime with Profile-Guided Optimizations</a> was originally published in <a href="https://proandroiddev.com/">ProAndroidDev</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>