<?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 Khyati Agarwal on Medium]]></title>
        <description><![CDATA[Stories by Khyati Agarwal on Medium]]></description>
        <link>https://medium.com/@KHYATI1?source=rss-fc00a7d4bbc0------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*7ODK0LsJ4oQAd6tu.</url>
            <title>Stories by Khyati Agarwal on Medium</title>
            <link>https://medium.com/@KHYATI1?source=rss-fc00a7d4bbc0------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 23:15:32 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@KHYATI1/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[MVD — Minimum Viavl]]></title>
            <link>https://medium.com/@KHYATI1/mvd-minimum-viavl-e8a62868bcdd?source=rss-fc00a7d4bbc0------2</link>
            <guid isPermaLink="false">https://medium.com/p/e8a62868bcdd</guid>
            <category><![CDATA[mvd]]></category>
            <dc:creator><![CDATA[Khyati Agarwal]]></dc:creator>
            <pubDate>Mon, 14 Jul 2025 10:37:10 GMT</pubDate>
            <atom:updated>2025-07-14T13:46:47.032Z</atom:updated>
            <content:encoded><![CDATA[<h3>MVD — Minimum Viable Data</h3><p>MVP doesn’t work for AI and data products. So, AI and data product managers should focus on AI MVD — <strong>Minimum Viable Data — </strong><em>the smallest amount of data required to develop an AI Product.</em></p><p>Rather than working with very large data sets that take time to store, clean and create pipelines for ai product, MVD starts with the data science team starting with the smallest amount of data required to build an AI Product. An MVD is not about creating a neural network and with time adding more and more layers to it, rather it is, about methods that require very little data and delivering to the user, and overtime as the complexity increases, replacing the model with another model that uses just the right amount of data(a little bit more data), before finally needing mass amounts of data. If one is developing image recognition system or a robot that learns to walk, one has no choice but to leverage a complex model that uses a large amount of data. So there are exceptions to the MVD method. However, the principle still applies. So, the MVD helps businesses in:</p><p>leveraging the right data, get better performance through MVD iteration, the results in each MVD iteration provide some value to us or the user. Through the MVD, it helps in providing less time to storing, cleaning and applying ML model to the data, thus ensuring minimal data processing required, it is cost effective and is time efficient.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e8a62868bcdd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[MVP — Minimum Viable Product]]></title>
            <link>https://medium.com/@KHYATI1/mvp-minimum-viable-product-54831f7afa45?source=rss-fc00a7d4bbc0------2</link>
            <guid isPermaLink="false">https://medium.com/p/54831f7afa45</guid>
            <category><![CDATA[mvp]]></category>
            <dc:creator><![CDATA[Khyati Agarwal]]></dc:creator>
            <pubDate>Mon, 14 Jul 2025 10:36:27 GMT</pubDate>
            <atom:updated>2025-07-14T10:36:27.530Z</atom:updated>
            <content:encoded><![CDATA[<h3>MVP — Minimum Viable Product</h3><p>MVP is the simplest version of the product solution. The goal is to release the minimalist version of the product to the customers so that feedback can be collected in a low effort way and integrated into the product’s next iteration. With MVP, the goal is not to over invest in the ideas and waste resources only to understand whether the feature/product is required or not. Essentially we are looking for the lowest effort that is done in order to collect feedback from the customer before continuing to build out the product features.</p><p>The common metaphor of an MVP is that of building a car. Traditionally building a car would mean building each of its complex components first and then integrating them to build the car. However, in the modern agile approach, the goal is to see for who are we building the product, what purpose it solves. The purpose it solves is to move from point A to point B. Thus indicating that instead of a car, the purpose can be accomplished by maybe a skate board or a cycle. We can get the user to start using the product and get the feedback. With each iteration, the product turns into something new, but with each iteration we are developing something the customer wants sooner and not something which is big and the user might not want. Its a less risking way to build a product. That is an MVP. That is how we should develop a product as an AI and data product manager. <em>Take the product that we are looking to build and pair it down to the simplest working version that could be launched with users before developing the idea further or scraping the idea altogether.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=54831f7afa45" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Product Ideation Techniques apart from Brainstorming]]></title>
            <link>https://medium.com/@KHYATI1/product-ideation-techniques-apart-from-brainstorming-466060545002?source=rss-fc00a7d4bbc0------2</link>
            <guid isPermaLink="false">https://medium.com/p/466060545002</guid>
            <category><![CDATA[ideation]]></category>
            <dc:creator><![CDATA[Khyati Agarwal]]></dc:creator>
            <pubDate>Wed, 09 Jul 2025 08:44:45 GMT</pubDate>
            <atom:updated>2025-07-09T08:53:11.756Z</atom:updated>
            <content:encoded><![CDATA[<p>Take a pen, a piece of paper and preferably group of people need to be there who can do ideation. For all of the methods described below, take a <strong>variety</strong> of <strong>different</strong> people who can take part in it from the company.</p><p>The 3 methods are- Crazy Eights, Round Robin and The Fake Press Release</p><p>Crazy Eights — Fold the piece of paper multiple times to get 8 sections of the paper. Each person needs to do this. Now take the core problem/core dataset available and give 15 seconds for each section for ideation. The idea is to give so little time for ideation that individuals wont be embarrassed to throw first idea that comes to mind.</p><p>Round Robin — Everyone in the group is given a piece of paper and they are given a minute to write an idea, once that time passes, they give their paper to someone on the left and take the paper from someone on the right, read the idea and then ideate over this idea for another minute. This is repeated until their own page comes back to them. The benefit of this method is that the idea gets further developed from the core idea and that it engages people all throughout the group to offer ideas.</p><p>The Fake Press Release — Here, take the core problem or the core data set and ask the team to write the fake press release for the idea that will solve the core problem or leverage the data set under consideration. Everyone is given 5 minutes to do this. Do not write the entire press release but write out the following — idea summary, press headline and customer quote. Since press loves to pick on big and exciting ideas so this gives the opportunity to think big.</p><p>Brainstorming is done top down and bottom up, so need to leverage many ideation techniques and as many people as possible as a data and ai product manager.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=466060545002" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What is AI Flywheel Effect]]></title>
            <link>https://medium.com/@KHYATI1/what-is-ai-flywheel-effect-3af7d5b4f670?source=rss-fc00a7d4bbc0------2</link>
            <guid isPermaLink="false">https://medium.com/p/3af7d5b4f670</guid>
            <dc:creator><![CDATA[Khyati Agarwal]]></dc:creator>
            <pubDate>Mon, 19 May 2025 12:11:37 GMT</pubDate>
            <atom:updated>2025-05-19T12:15:05.689Z</atom:updated>
            <content:encoded><![CDATA[<p>AI products operate in a virtual cycle called the AI Flywheel Effect. Instead of ai products declining overtime, they are improved overtime (more and more competitive overtime). For instance, a tech like image recognition does not diminish in value overtime, but through feedback loops gets betters and more useful overtime.</p><p>The AI Flywheel looks like this: with data you can build AI to make predictions, with better predictions one has better user experience, with better experience there are more users also known as traffic, which in turn results in more data and so on.</p><p>Data → Predictions → Customer Experience → Traffic → Data (Circular loop)</p><p>For example, in medium app we can see, when we are writing our blog, we are able to get recommendations of the next word or a phrase that we might need, so that, not every word needs to be typed, this uses AI autocomplete beneath for better recommendations which right now, i have used, in my scenario. This is an example of better user experience which lead to more users including me using it, leads to more data and in turn gives us better predictions and so on.</p><p>In essence, an AI product becomes more competitive overtime, whereas other technological products deprecate overtime due to being less competitive overtime.</p><p>This means, as an ai product manager, we do need to ensure that our product model fits to this cyclical effect to generate more traffic, which results in generating more data which makes our ai model more competitive and so on.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3af7d5b4f670" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Concept behind open source]]></title>
            <link>https://medium.com/@KHYATI1/concept-behind-open-source-a08c5a5ef9d5?source=rss-fc00a7d4bbc0------2</link>
            <guid isPermaLink="false">https://medium.com/p/a08c5a5ef9d5</guid>
            <category><![CDATA[open-source-development]]></category>
            <dc:creator><![CDATA[Khyati Agarwal]]></dc:creator>
            <pubDate>Wed, 23 Apr 2025 15:56:15 GMT</pubDate>
            <atom:updated>2025-04-23T15:56:15.348Z</atom:updated>
            <content:encoded><![CDATA[<p>The <strong>concept behind open source</strong> revolves around the idea of <strong>transparency, collaboration, and freedom</strong> in software development. Here’s a breakdown of what that means:</p><h3>🔓 1. Open Access to Source Code</h3><ul><li>The <strong>source code</strong> of the software is made publicly available.</li><li>Anyone can view, study, and learn from it.</li><li>Unlike proprietary software (e.g., Microsoft Office), which hides its internals, open source invites transparency.</li></ul><h3>🤝 2. Freedom to Modify and Share</h3><ul><li>Users can <strong>modify the code</strong> to fit their needs.</li><li>They can <strong>redistribute</strong> original or modified versions.</li><li>This encourages <strong>innovation</strong> and <strong>community improvements</strong>.</li></ul><h3>🌍 3. Community Collaboration</h3><ul><li>Developers from around the world contribute to projects.</li><li>Think of it like Wikipedia but for code.</li><li>Popular examples: <strong>Linux</strong>, <strong>Python</strong>, <strong>Firefox</strong>, <strong>Kubernetes</strong>, <strong>TensorFlow</strong>.</li></ul><h3>📜 4. Licenses Govern Usage</h3><ul><li>Open source licenses (like <strong>MIT</strong>, <strong>GPL</strong>, or <strong>Apache</strong>) define what you’re allowed to do.</li><li>Some are very permissive (MIT), others ensure modifications stay open (GPL).</li></ul><h3>🚀 5. Benefits</h3><ul><li><strong>Faster development</strong> through global collaboration.</li><li><strong>Fewer bugs</strong>, as more eyes review the code.</li><li><strong>Lower costs</strong>, especially for startups or researchers.</li><li>Encourages <strong>learning</strong> and <strong>skills development</strong>.</li></ul><p>If you’re thinking about contributing to or using open source, it’s not just about free software — it’s about being part of a global, evolving ecosystem of innovation.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a08c5a5ef9d5" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What is Big data]]></title>
            <link>https://medium.com/@KHYATI1/what-is-big-data-a0ed936164fa?source=rss-fc00a7d4bbc0------2</link>
            <guid isPermaLink="false">https://medium.com/p/a0ed936164fa</guid>
            <category><![CDATA[big-data]]></category>
            <dc:creator><![CDATA[Khyati Agarwal]]></dc:creator>
            <pubDate>Wed, 23 Apr 2025 09:46:21 GMT</pubDate>
            <atom:updated>2025-04-23T09:52:14.309Z</atom:updated>
            <content:encoded><![CDATA[<p>A dataset too big to be processed on a single machine can be termed as big data. Most people say, big data is huge terrabytes of data, but small data is also processed by hadoop successfully. So a reasonable definition of big data is the one mentioned above.</p><p>There are 3Vs to the big data- volume, velocity and variety</p><p>Volume — — size of big data</p><p>velocity — speed at which the data set grows</p><p>variety —the different data formats coming from same or different data sources.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a0ed936164fa" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Difference between machine learning and traditional software development]]></title>
            <link>https://medium.com/@KHYATI1/difference-between-machine-learning-and-traditional-software-development-97373805dffd?source=rss-fc00a7d4bbc0------2</link>
            <guid isPermaLink="false">https://medium.com/p/97373805dffd</guid>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Khyati Agarwal]]></dc:creator>
            <pubDate>Wed, 23 Apr 2025 09:19:09 GMT</pubDate>
            <atom:updated>2025-04-23T09:31:04.858Z</atom:updated>
            <content:encoded><![CDATA[<p>In case of machine learning, we know the input and the output but we do not know the algorithm whereas in case of traditional software development, we know the input and the algorithm and we use that to give the output.</p><p>So, in traditional software development, the algorithm is known whereas in machine learning, the input and output are known. In case of machine learning, we use neural networks to generate a relationship between the inputs and the outputs. A neural network can be thought of as a stack of layer where each layer consists of some predefined math and internal variables. In other words, a machine learning algorithm can be defined as an algorithm that uses some internal variables in stacks of layer that finee tunes the input to generate some output.</p><p>The input here is called the feature whereas the output is called the label. The training happens by feeding the pair of input and output to the machine learning model which has been used in any case. That is then called a training example.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=97373805dffd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Design patterns]]></title>
            <link>https://medium.com/@KHYATI1/design-patterns-5e34fa2c5240?source=rss-fc00a7d4bbc0------2</link>
            <guid isPermaLink="false">https://medium.com/p/5e34fa2c5240</guid>
            <category><![CDATA[design-patterns]]></category>
            <dc:creator><![CDATA[Khyati Agarwal]]></dc:creator>
            <pubDate>Mon, 14 Apr 2025 09:56:08 GMT</pubDate>
            <atom:updated>2025-04-14T09:57:43.789Z</atom:updated>
            <content:encoded><![CDATA[<p>What are design patterns:</p><p>Design patterns are solutions to problems in software design. The aim is not to memorise the syntax, but develop the ability to solve the problems using these design patterns. There are 23 different design patterns broadly categorised into 3 categories</p><ol><li>Structural design patterns</li><li>Behavioural design patterns</li><li>Creational design patterns</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5e34fa2c5240" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What is MCP — Source ByteByteGo]]></title>
            <link>https://medium.com/@KHYATI1/what-is-mcp-source-bytebytego-0133ec3cee05?source=rss-fc00a7d4bbc0------2</link>
            <guid isPermaLink="false">https://medium.com/p/0133ec3cee05</guid>
            <category><![CDATA[mcp-server]]></category>
            <dc:creator><![CDATA[Khyati Agarwal]]></dc:creator>
            <pubDate>Wed, 09 Apr 2025 06:58:37 GMT</pubDate>
            <atom:updated>2025-04-09T06:58:37.304Z</atom:updated>
            <content:encoded><![CDATA[<h3>What is MCP — Source ByteByteGo</h3><p>MCP — — Model context protocol is an open standard that enables seamless integration between ai models and Claude an external data sources, replacing fragmented integration with a single data source. AI systems can thus be given access to databases, file systems, apis and other tools in a standardised way.</p><p>Architecture: It follows a client server model with 3 key components — hosts, clients and servers.</p><p>Servers support 3 primitives — prompts, resources and tools. Prompts are instructions or templates which guide how the LLM should approach certain tasks with data. Tools are executable functions that the LLM can call to retrieve information or perform actions outside its context like querying a database or modifying the file.</p><p>Clients have 2 primitives — roots and sampling. The clients and servers are the building blocks that enable standardised communication between ai models and external systems.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*q1yZd8x5tJzpQgw9-tW_Nw.png" /><figcaption>source — ByteByteGo</figcaption></figure><p>As we can see from the above diagram, MCP simplifies the integration landscape with Claude being a practical example.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0133ec3cee05" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The lines between testing and software engineering have blurred.]]></title>
            <link>https://medium.com/@KHYATI1/the-lines-between-testing-and-software-engineering-have-blurred-ccef307d901c?source=rss-fc00a7d4bbc0------2</link>
            <guid isPermaLink="false">https://medium.com/p/ccef307d901c</guid>
            <dc:creator><![CDATA[Khyati Agarwal]]></dc:creator>
            <pubDate>Wed, 13 Nov 2019 16:44:51 GMT</pubDate>
            <atom:updated>2019-11-14T13:33:12.812Z</atom:updated>
            <content:encoded><![CDATA[<p>The lines between testing and software engineering have blurred.</p><p>With increasing automation, the lines between testing and development have blurred. Where developers write code to test the application, software design engineers in test write code to test the application. This means both have to be equivalently good in coding, to ensure proper delivery of the product to be available in the market where multiple services are being provided to the customers.</p><p>While not only does this mean much much relief to QA from the mundane tasks of manual test, automation still requires some coding ability and speed, keeping pace with the organisational demands to succeed in the market. This is something that might prove to be tedious to some who do not like the job whereas upliftment to some who do not like manual work. Automation scripts can be deployed and run in the environment. While this is a good news. What is daunting is that once this skill set is met with the demand, once the work is accomplished, people would still be sitting idle or require continuous technological upgrade to keep pace with the changing business.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ccef307d901c" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>