<?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 Rohanvishwakarma on Medium]]></title>
        <description><![CDATA[Stories by Rohanvishwakarma on Medium]]></description>
        <link>https://medium.com/@rohanvishwakarma182?source=rss-3ccb380f9c9d------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*vRzhAd57VB8tt6QA.jpg</url>
            <title>Stories by Rohanvishwakarma on Medium</title>
            <link>https://medium.com/@rohanvishwakarma182?source=rss-3ccb380f9c9d------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 01:48:44 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@rohanvishwakarma182/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[Are We Really Getting Faster at Software Development?]]></title>
            <link>https://medium.com/@rohanvishwakarma182/are-we-really-getting-faster-at-software-development-e756973ef76a?source=rss-3ccb380f9c9d------2</link>
            <guid isPermaLink="false">https://medium.com/p/e756973ef76a</guid>
            <dc:creator><![CDATA[Rohanvishwakarma]]></dc:creator>
            <pubDate>Fri, 02 Oct 2026 17:04:26 GMT</pubDate>
            <atom:updated>2026-10-02T17:04:26.396Z</atom:updated>
            <content:encoded><![CDATA[<p>Are We Really Getting Faster at Software Development?</p><p>Just a thought from the day-to-day life of a developer.</p><p>This isn&#39;t a guide, a prediction, or a &quot;10 things every developer should do&quot; post.</p><p>It&#39;s simply something I&#39;ve been noticing while working on software.</p><p>We have better tools than ever.</p><p>Faster frameworks.</p><p>Better cloud platforms.</p><p>Automation everywhere.</p><p>And now, AI tools like Claude can help us write, debug, refactor, and understand code much faster.</p><p>So naturally, it feels like software development should be getting easier.</p><p>But sometimes, I feel we&#39;re experiencing something slightly different.</p><p>We&#39;re getting faster at creating software, but not always faster at maintaining it.</p><p>And I think that&#39;s an interesting distinction.</p><p>---</p><p>🚀 In the Beginning, Everything Feels Fast</p><p>When you&#39;re starting a new project, things are usually straightforward.</p><p>You have a relatively small codebase.</p><p>The architecture is easier to understand.</p><p>There aren&#39;t hundreds of dependencies.</p><p>A developer can make a change, test it, and move on.</p><p>You might think:</p><p>«&quot;Why spend so much time designing this? We can just build it and improve it later.&quot;»</p><p>And sometimes, that&#39;s completely reasonable.</p><p>Early-stage development benefits from moving quickly.</p><p>The problem starts when the project grows but the way we build doesn&#39;t evolve with it.</p><p>---</p><p>🧱 Small Decisions Start Adding Up</p><p>I&#39;ve seen how a small shortcut can feel completely harmless when you&#39;re taking it.</p><p>Maybe there&#39;s no time to refactor.</p><p>So you copy a piece of logic.</p><p>Maybe the architecture isn&#39;t clear yet.</p><p>So you put the functionality wherever it seems convenient.</p><p>Maybe there&#39;s no time to write tests.</p><p>So you tell yourself you&#39;ll add them later.</p><p>Maybe nobody documents why something was built that way.</p><p>Because everyone currently working on it already knows.</p><p>Nothing seems particularly wrong.</p><p>Until six months later.</p><p>That copied logic exists in multiple places.</p><p>That temporary solution is now depended upon by several features.</p><p>Nobody remembers why a particular decision was made.</p><p>And suddenly, changing one small thing requires understanding half the application.</p><p>---</p><p>⏳ The Strange Part: The Team Isn&#39;t Necessarily Slower</p><p>This is what I find interesting.</p><p>The developers haven&#39;t suddenly become worse.</p><p>They&#39;re still working hard.</p><p>The tools are better.</p><p>The development environment is better.</p><p>But the cost of changing the system has increased.</p><p>A change that once took a few hours now takes a couple of days.</p><p>Not because writing the code takes longer.</p><p>Because understanding the consequences takes longer.</p><p>You have to ask:</p><p>- What else depends on this?<br>- Will this break another service?<br>- Is this logic duplicated somewhere?<br>- Which database records are affected?<br>- Is there an old workaround we need to preserve?<br>- Which tests cover this behavior?<br>- Why was this originally implemented this way?</p><p>At some point, understanding becomes more expensive than writing.</p><p>---</p><p>🤖 Then AI Changed the Equation</p><p>And now we have another interesting development.</p><p>Tools like Claude can generate code incredibly quickly.</p><p>You can describe a requirement and get an implementation.</p><p>Ask for a refactor, and you can get one.</p><p>Ask for tests, and you can generate them.</p><p>Ask it to explain an unfamiliar function, and you can get a useful starting point.</p><p>As a developer, that&#39;s genuinely useful.</p><p>But it also made me think about something:</p><p>«If producing code becomes dramatically cheaper, does maintaining all that code become cheaper too?»</p><p>Not necessarily.</p><p>AI can help us produce code faster.</p><p>But the system still has to live with that code.</p><p>Someone still has to review it.</p><p>Someone has to understand the architecture.</p><p>Someone has to test the edge cases.</p><p>Someone has to debug it when real users do something the original prompt didn&#39;t anticipate.</p><p>So perhaps the bottleneck is slowly moving.</p><p>Before:</p><p>Writing code was expensive.</p><p>Now:</p><p>Understanding, validating, integrating, and maintaining code can become the expensive part.</p><p>---</p><p>🔥 When &quot;Moving Fast&quot; Turns Into Constant Fixing</p><p>Another pattern I&#39;ve noticed is how quickly development can become reactive.</p><p>A feature gets shipped.</p><p>Then there&#39;s a bug.</p><p>A quick fix goes out.</p><p>That fix creates another issue.</p><p>Another patch follows.</p><p>Meanwhile, the next feature is already waiting.</p><p>So the team keeps moving.</p><p>But not necessarily forward.</p><p>The backlog grows.</p><p>Technical debt grows.</p><p>The number of things developers need to keep in their heads grows.</p><p>And eventually you hear something like:</p><p>«&quot;Don&#39;t touch that part. Nobody knows what will happen.&quot;»</p><p>That&#39;s probably one of the clearest signs that a system has become difficult to change.</p><p>---</p><p>⚖️ I&#39;m Not Saying We Should Slow Down</p><p>That&#39;s not the point.</p><p>We shouldn&#39;t turn every small feature into a six-week architecture exercise.</p><p>We shouldn&#39;t document every line of code.</p><p>We shouldn&#39;t avoid shortcuts completely.</p><p>And we shouldn&#39;t stop using AI because generated code can introduce complexity.</p><p>Speed is valuable.</p><p>AI is valuable.</p><p>Rapid experimentation is valuable.</p><p>The question is about balance.</p><p>Maybe the real skill is knowing where to move quickly and where to deliberately slow down.</p><p>Move quickly when you&#39;re exploring.</p><p>Move quickly when the cost of being wrong is small.</p><p>But when you&#39;re changing a core system that dozens of future features will depend on, perhaps spending extra time thinking isn&#39;t slowing development down.</p><p>Maybe it&#39;s protecting future development speed.</p><p>---</p><p>🧠 Something I&#39;ve Started Thinking About</p><p>Instead of asking only:</p><p>«&quot;How quickly can we build this?&quot;»</p><p>Maybe we should also ask:</p><p>«&quot;How easy will this system be to change after we&#39;ve built it?&quot;»</p><p>Because the real cost of software isn&#39;t always the first implementation.</p><p>It&#39;s every change that comes after it.</p><p>If today&#39;s implementation makes tomorrow&#39;s changes easier, that&#39;s an investment.</p><p>If today&#39;s shortcut makes every future change harder, we&#39;re borrowing time from the future.</p><p>And eventually, that bill arrives.</p><p>---</p><p>👨‍💻 Just a Developer&#39;s Perspective</p><p>Again, this isn&#39;t meant to be a rule for everyone.</p><p>Different teams have different constraints.</p><p>A startup trying to validate an idea has different priorities from a team maintaining a financial system.</p><p>A prototype can tolerate things that production infrastructure cannot.</p><p>And sometimes a shortcut is exactly the right decision.</p><p>I&#39;m simply sharing something I&#39;ve noticed in day-to-day development:</p><p>The faster we become at producing code, the more important it becomes to think about the code we choose to keep.</p><p>AI makes that conversation even more interesting.</p><p>Because when generating 500 lines of code becomes easy, the valuable question may no longer be:</p><p>&quot;Can we build it?&quot;</p><p>It may be:</p><p>&quot;Should we build it this way?&quot;</p><p>---</p><p>💬 What Do You Think?</p><p>This is just my perspective from working with software and seeing how projects evolve over time.</p><p>I&#39;m genuinely curious about other developers&#39; experiences.</p><p>Have you noticed this too?</p><p>Does AI actually make your development process faster end-to-end, or does it sometimes just move the bottleneck from writing code → reviewing, understanding, testing, and maintaining it?</p><p>And what&#39;s been the biggest source of slowdown in your projects?</p><p>Technical debt? Architecture? Changing requirements? Too many dependencies? AI-generated code? Something else?</p><p>I&#39;d genuinely like to hear different experiences.</p><p>What&#39;s your take?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e756973ef76a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Top 5 basic function that used in Pytorch library]]></title>
            <link>https://medium.com/@rohanvishwakarma182/top-5-basic-function-that-used-in-pytorch-library-be29a4ada572?source=rss-3ccb380f9c9d------2</link>
            <guid isPermaLink="false">https://medium.com/p/be29a4ada572</guid>
            <dc:creator><![CDATA[Rohanvishwakarma]]></dc:creator>
            <pubDate>Sat, 30 May 2020 15:37:59 GMT</pubDate>
            <atom:updated>2020-05-30T15:37:59.414Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*zaP63YeygYUrthQE.png" /></figure><h3>Roadmap for the post</h3><ol><li>A brief introduction to PyTorch</li><li>getting started with pytorch</li><li>torch.is_leaf()</li><li>torch.split()</li><li>torch.zeros()</li><li>torch.fill_diagonal()</li><li>torch.reciprocal()</li></ol><h3>What is PyTorch?</h3><blockquote>PyTorch is an open-source machine learning library for Python which allows maximum flexibility and speed on scientific computing for deep learning. It is a replacement for NumPy to use the power of GPUs . More want to know about pytorch</blockquote><blockquote><a href="https://pytorch.org/">https://pytorch.org/</a></blockquote><h3>Getting started with pytorch?</h3><p>Before you get started with code, you need to install the latest version of PyTorch. We are using Google Colab for our tutorial, so we will use the following command to install PyTorch.</p><pre>!pip3 install torch==1.2.0+cu92 torchvision==0.4.0+cu92 -f https://download.pytorch.org/whl/torch_stable.html</pre><p>Now we need to import a few modules which will be useful to obtain the necessary functions that will help us to build our neural network model. The main ones are torch and torchvision. They contain the majority of the functions that you need to get started with PyTorch. However, we will need torch.nn, torch.nn.functional and torchvision.transforms which all contain utility functions to build our model. We probably won&#39;t use all the time time but this are most important to build your own project.</p><p>Below we check for the PyTorch version just to make sure that you are using the proper version that was used for this tutorial. At the time of this tutorial, we are working with PyTorch 1.2.</p><pre>print(torch.__version__)</pre><p>A tensor is an n-dimensional data container which is similar to NumPy’s ndarray. For example, 1d-tensor is a vector, 2d-tensor is a matrix, 3d-tensor is a cube, and 4d-tensor is a vector of cubes.</p><p>Let’s take a look at some examples of how to create a tensor in PyTorch. To Initialize a tensor, we can either assign values directly or set the size of the tensor. torch.Tensor(n,m) will initialize a tensor with size n x m.</p><h3>Installation process of pytorch in jupyter notebook</h3><p>In [1]: <strong>!</strong>pip install torch<strong>==</strong>1.4.0<strong>+</strong>cpu torchvision<strong>==</strong>0.5.0<strong>+</strong>cpu <strong>-</strong>f <a href="https://download.pytorch.org/whl/torch_stable.html">https:<strong>//</strong>download.pytorch.org<strong>/</strong>whl<strong>/</strong>torch_stable.html</a></p><pre> Looking in links: <a href="https://download.pytorch.org/whl/torch_stable.html">https://download.pytorch.org/whl/torch_stable.html</a><br>Requirement already satisfied: torch==1.4.0+cpu in c:\users\hp\appdata\local\programs\python\python37\lib\site-packages (1.4.0+cpu)<br>Requirement already satisfied: torchvision==0.5.0+cpu in c:\users\hp\appdata\local\programs\python\python37\lib\site-packages (0.5.0+cpu)<br>Requirement already satisfied: numpy in c:\users\hp\appdata\local\programs\python\python37\lib\site-packages (from torchvision==0.5.0+cpu) (1.17.1)<br>Requirement already satisfied: six in c:\users\hp\appdata\local\programs\python\python37\lib\site-packages (from torchvision==0.5.0+cpu) (1.13.0)<br>Requirement already satisfied: pillow&gt;=4.1.1 in c:\users\hp\appdata\local\programs\python\python37\lib\site-packages (from torchvision==0.5.0+cpu) (6.1.0)</pre><p>above command is used install pytorch libarry</p><p>In [ ]:</p><pre><em># Import torch and other required modules</em></pre><pre><strong>import</strong> torch</pre><h3>torch.is_leaf()</h3><p>All Tensors that have requires_grad which is False will be leaf Tensors by convention.</p><p>For Tensors that have requires_grad which is True, they will be leaf Tensors if they were created by the user. This means that they are not the result of an operation and so grad_fn is None.</p><p>Only leaf Tensors will have their grad populated during a call to backward(). To get grad populated for non-leaf Tensors, you can use retain_grad().</p><p>In [3]:</p><pre><em># Example 1</em></pre><pre>a <strong>=</strong> torch.rand(10, requires_grad<strong>=True</strong>)<br>a.is_leaf</pre><p>Out[3]:</p><pre>True</pre><pre>Here, a is created by user input. That’s why leaf value is true</pre><ul><li>In [6]:</li></ul><pre><em># Example 2<br></em>c <strong>=</strong> torch.rand(10, requires_grad<strong>=True</strong>) <strong>+</strong> 2<br>c.is_leaf​</pre><pre>Here leaf value <strong>is</strong> false because c <strong>is</strong> created by addition operation</pre><p>Out[6]:</p><pre>False</pre><h3>usecase: torch.is_leaf should be used to check if the tensor was created by user or other operations</h3><h3>torch.split()</h3><p>torch.split takes “size” of chunk/chunks and split it into required sections</p><p>In [14]:</p><pre><em># Example 1 - working</em></pre><pre>a <strong>=</strong> torch.randn(10, 10) <em>#tensor of size 50 x 80</em></pre><pre>b <strong>=</strong> torch.split(a, 5, dim<strong>=</strong>0) <em># it returns a tuple</em></pre><pre>b <strong>=</strong> list(b) <em># convert to list if you want</em></pre><pre>b</pre><p>Out[14]:</p><pre>[tensor([[-1.4230, -0.3360, -0.0537,  1.0705, -0.1434, -1.2120, -1.0587,  0.7570,<br>           1.1338,  1.0054],<br>         [ 0.7402, -1.1350, -0.0700, -0.5089, -0.0956,  0.2634, -0.2267,  0.9803,<br>          -0.3216, -0.5696],<br>         [-0.2491, -0.9374,  1.1733,  0.2544,  0.5445, -0.2122,  1.6761, -0.1776,<br>          -0.3317, -0.9594],<br>         [ 1.5069, -1.1113, -0.1766, -0.0328,  0.0380, -0.8963, -0.8364,  0.2324,<br>           0.1386,  0.5274],<br>         [-1.6306,  0.0074,  1.1587, -0.9156, -0.3634,  0.9089, -1.0933,  0.0724,<br>           1.0921, -1.4032]]),<br> tensor([[ 0.5762,  0.4257, -0.9031,  0.6173,  0.3570, -0.8923,  0.0105,  0.4663,<br>           1.1005, -0.5071],<br>         [-0.2397,  0.3719,  0.0801, -0.3463,  0.3342,  2.1502, -0.0556, -0.8743,<br>           0.4031,  0.3773],<br>         [ 2.3744, -0.8310, -0.0965,  1.6210,  0.1699,  0.5689,  1.3094, -0.5738,<br>          -1.3827, -0.6038],<br>         [-1.4591, -0.4414,  1.7072,  0.3172,  0.0726,  1.8749, -0.3376, -0.5245,<br>          -0.6642, -0.8985],<br>         [-0.6973,  0.7947,  0.5202,  1.1960,  2.0960, -0.4976, -0.2815,  0.1963,<br>           0.9078, -1.1606]])]</pre><pre>Here, we have created a random 10x10 tensor and put it in a. And we have split it with 5 sections with dim=1[note: dimenstion depends on the tensor size] and put it into b. Than we print b.</pre><p>In [15]:</p><pre><em># Example 2 — working</em></pre><pre>a <strong>=</strong> torch.randn(2, 2) <em>#tensor of size 50 x 80</em></pre><pre>b <strong>=</strong> torch.split(a, 2, dim<strong>=</strong>1) <em># it returns a tuple<br></em>b <strong>=</strong> list(b) <em># convert to list if you want</em></pre><p>Out[15]:</p><pre>[tensor([[-1.4915, -1.0299],<br>         [-1.2023,  0.5340]])]</pre><p>As you can see, we have 2x2 tensors. We took dim=1 and split it into 2 part.</p><h3>usecase: torch.splits — Splits the tensor into chunks. Each chunk is a view of the original tensor.</h3><h3>torch.zeros()</h3><p>we can get a matrix consisting of all zeros by using torch.zeros()</p><p>In [16]:</p><pre><em># Example 1 — working</em></pre><pre>a <strong>=</strong> torch.zeros((3,3))<br>print(a)<br>print(a.shape)</pre><pre>tensor([[0., 0., 0.],<br>        [0., 0., 0.],<br>        [0., 0., 0.]])<br>torch.Size([3, 3])</pre><p>Here we have created a 3x3 matrix with all zeros.</p><h3>torch.fill_diagonal()</h3><p>In [17]:</p><pre><em># Example 1 — working</em></pre><pre>a <strong>=</strong> torch.zeros(3, 3)</pre><pre>a.fill_diagonal_(5)</pre><p>Out[17]:</p><pre>tensor([[5., 0., 0.],<br>        [0., 5., 0.],<br>        [0., 0., 5.]])</pre><p>Here we have create a 3x3 matrix with 0. and filled the diagonal with 5.</p><p>In [18]:</p><pre><em># Example 2 — working</em></pre><pre>b <strong>=</strong> torch.zeros(7, 3)</pre><pre>b.fill_diagonal_(5, wrap<strong>=True</strong>)</pre><p>Out[18]:</p><pre>tensor([[5., 0., 0.],<br>        [0., 5., 0.],<br>        [0., 0., 5.],<br>        [0., 0., 0.],<br>        [5., 0., 0.],<br>        [0., 5., 0.],<br>        [0., 0., 5.]])</pre><h3>torch.reciprocal()</h3><pre>Returns a new tensor with the reciprocal of the elements of input.</pre><pre>output[i] = 1/input[i]<br>Parameters:<br>input (Tensor) — the input tensor.</pre><pre>out (Tensor, optional) — the output tensor.</pre><p>In [19]:</p><pre><em># Example 1 — working<br></em>a <strong>=</strong> torch.randn(4)<br>a</pre><p>Out[19]</p><pre>tensor([-1.2976, -0.1140, -1.1232, -0.8342])</pre><p>In [20]:</p><pre>torch.reciprocal(a)</pre><p>Out[20]:</p><pre>tensor([-0.7707, -8.7749, -0.8903, -1.1988])</pre><p>Here, torch.reciprocal returns 1/input[i].</p><h3>Conclusion</h3><pre>So far, we have covered in this post 5 different functions with example which is used in PyTorch in different case.</pre><h3>Reference Links</h3><pre><a href="https://pytorch.org/docs/stable/tensors.html">https://pytorch.org/docs/stable/tensors.html</a></pre><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=be29a4ada572" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>