<?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 Learn With Mohit Vyas on Medium]]></title>
        <description><![CDATA[Stories by Learn With Mohit Vyas on Medium]]></description>
        <link>https://medium.com/@learnwithmohitvyas?source=rss-5811f2e1bf2c------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*BVKSnEn8tKJwFKOH</url>
            <title>Stories by Learn With Mohit Vyas on Medium</title>
            <link>https://medium.com/@learnwithmohitvyas?source=rss-5811f2e1bf2c------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 08:16:39 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@learnwithmohitvyas/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[Limitations of Foundation models]]></title>
            <link>https://medium.com/@learnwithmohitvyas/limitations-of-foundation-models-6b7987a5cc2c?source=rss-5811f2e1bf2c------2</link>
            <guid isPermaLink="false">https://medium.com/p/6b7987a5cc2c</guid>
            <category><![CDATA[genai]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[deep-learning]]></category>
            <dc:creator><![CDATA[Learn With Mohit Vyas]]></dc:creator>
            <pubDate>Fri, 13 Mar 2026 03:27:23 GMT</pubDate>
            <atom:updated>2026-03-13T03:27:23.054Z</atom:updated>
            <content:encoded><![CDATA[<p>Foundation models, which are the backbone of generative AI, are very powerful tools for implementing the generative AI. But these powerful tools do not come without their limitations. So today we will be learning about what are the limitations of these foundation models. Understanding and knowing these limitations will help us in a better and effective utilization of these powerful tools.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*6D_1nCb_ViCJmSec" /><figcaption>Photo by <a href="https://unsplash.com/@omilaev?utm_source=medium&amp;utm_medium=referral">Igor Omilaev</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>The first limitation is data dependency. The foundation models are heavily dependent on the data that is fed into them. Any inconsistency or incompleteness in the data that is fed into the foundation models will be clearly visible in the output that is generated by these foundation models in the form of generative AI response. Asking generative ai model or the foundation model about something which has not been fed into it is similar to asking a student to write an essay on a book which he has not read.</p><p>Second limitation is Knowledge cut-off. It is the last date that an AI model was trained on new information. If you remember the early days of ChatGPT rollout, there used to be a disclaimer saying that this model is trained on the data only up to 31st Dec, 2021 or 2022. This limitation can make your model not in sync with the real-world happenings of the current date, as it is being fed the information which was 2 to 3 years ago. So once you prompt your model for any recent happenings in the world, it might not be able to give you the latest response.</p><p>Third limitation is bias. Bias can be defined as unbalanced information in the large amount of data that is fed into the foundation model. Sometimes these biases in the data can be amplified by the foundation model and it can result into the response which is not in sync with the real truth.</p><p>Fourth limitation is fairness. Fairness in itself is a very subjective and a complex term. What might be fair for you might not be fair for me, or the other way around. But, in terms of foundation model, even if it is fed with completely balanced data, it might result in the output which is not completely fair. There are multiple fairness assessments of the generative AI model outputs, which categorize the output on different benchmarks of fairness. I would say they are not completely accurate all the time, and still improvement is needed on those assessments to achieve the goal of equitable AI.</p><p>Fifth limitation is hallucination. Hallucination means the model is giving output which is completely wrong, but it is giving that output in a way which sounds very convincing. Hallucination of the foundation models can be a deal breaker for the applications which are relying on the generative AI output for its accuracy. Hallucination needs a separate article which we would cover in detail in coming days.</p><p>Sixth limitation would be edge cases. So what are edge cases? So it is nothing but the rare and atypical scenarios that can expose the weaknesses of the foundation model, which might result in errors, misinterpretations, and the results which are not expected from the foundation models.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6b7987a5cc2c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Choosing a model for your Generative AI usecase]]></title>
            <link>https://medium.com/@learnwithmohitvyas/choosing-a-model-for-your-generative-ai-usecase-38dbac17fec0?source=rss-5811f2e1bf2c------2</link>
            <guid isPermaLink="false">https://medium.com/p/38dbac17fec0</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[genaimodels]]></category>
            <dc:creator><![CDATA[Learn With Mohit Vyas]]></dc:creator>
            <pubDate>Tue, 10 Mar 2026 05:27:51 GMT</pubDate>
            <atom:updated>2026-03-10T05:27:51.482Z</atom:updated>
            <content:encoded><![CDATA[<p>Factors to keep in mind while choosing a model for your GenAI use case. There are some criterias which are to be taken into consideration before choosing your model.</p><figure><img alt="Hold My Hand" src="https://cdn-images-1.medium.com/max/1024/0*HbczhoNJd6pu3xuJ" /><figcaption>Photo by <a href="https://unsplash.com/@cashmacanaya?utm_source=medium&amp;utm_medium=referral">Cash Macanaya</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Factor one is modality. Modality refers to the type of data the model can process and generate, such as text, images, videos, or audio. If your application focuses on a single data type, like generating text based on the articles or creating audio files, you’ll choose a model optimized for that specific modality. But for applications which require handling multiple data types, such as generating images, creating a video with accompanying audio, you will need a multimodal model.</p><p>Factor two would be context window. Context window refers to the amount of information a model can consider at one time before generating a response. The larger the context size, more the model can remember and generate the output related to that information. However, large context windows often require more computational cost, so you need to balance your context window based on your resource limitations.</p><p>Factor three would be security, which is of paramount importance, especially when dealing with sensitive data. You should consider the security features of the model, like data encryption, access controls, and the vulnerability management. Ensure that the model complies with the latest security standards and regulations as per your industry.</p><p>Factor four would be availability and reliability. So availability and reliability of the model are crucial for the production applications. We should choose a model that is consistently available and performs reliably under production environments. Consider factors like uptime guarantees, redundancy, and disaster recovery mechanisms.</p><p>Factor five would be cost, which is a significant factor in deciding the type of model that you want to choose. Different generative AI models can vary significantly in cost, so we should consider the pricing of the model, which might be based on usage, compute time, and the other factors. Evaluate the cost with respect to your budget and the expected value of the application. Bigger isn’t always better, so multimodal capabilities aren’t always necessary. So decide which features you want in your model and choose accordingly.</p><p>Factor six would be performance, which includes the accuracy, speed, and efficiency of the model. Evaluate the performance of the model on relevant benchmarks and datasets. Consider performance and the cost of the model for a better trade-off.</p><p>Factor seven would be fine-tuning and customization. Some models can be fine-tuned or customized for specific tasks or domains. If you are dealing with a specific use case, then consider a model that offers fine-tuning capabilities. This might often involve training a model further on a specific dataset related to your use case.</p><p>Factor eight would be ease of integration. So the most important thing is once you are done choosing the model, the chosen model should be easily integratable with your existing systems and workflows. So look for models which are having a well-documented APIs and SDKs for its easy integration into your existing system.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=38dbac17fec0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What is Deep learning? Where does it fit into the landscape of gen AI?]]></title>
            <link>https://medium.com/@learnwithmohitvyas/what-is-deep-learning-where-does-it-fit-into-the-landscape-of-gen-ai-0438bdeab5fd?source=rss-5811f2e1bf2c------2</link>
            <guid isPermaLink="false">https://medium.com/p/0438bdeab5fd</guid>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[neural-networks]]></category>
            <dc:creator><![CDATA[Learn With Mohit Vyas]]></dc:creator>
            <pubDate>Mon, 09 Mar 2026 04:24:04 GMT</pubDate>
            <atom:updated>2026-03-09T04:24:04.372Z</atom:updated>
            <content:encoded><![CDATA[<p>Recently, I was trying to learn about deep learning, as I’ve been listening about this topic from quite some time now.</p><figure><img alt="White robot human features" src="https://cdn-images-1.medium.com/max/1024/0*igw3q5R7xm3I68kr" /><figcaption>Photo by <a href="https://unsplash.com/@agk42?utm_source=medium&amp;utm_medium=referral">Alex Knight</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>First comes AI, then comes machine learning, then comes generative AI. AI is the parent, machine learning is the child, and the generative AI is the grandchild. You will be curious to know where does deep learning fits into this whole landscape of AI. Before that, first we’ll learn about what is deep learning and then we’ll dive into the landscape where deep learning fits into the whole picture.</p><p>What is deep learning? Deep learning is a powerful subset of machine learning, which is distinguished by its use of artificial neural networks. These neural networks enable the processing of highly complex patterns and generation of sophisticated predictions. Neural networks uses both labelled and unlabelled data, a strategy which is known as semi-supervised learning. They train on a blend of small amount of labelled data and large amount of unlabelled data. That way they learn the foundational concepts and generalize effectively to the new examples.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Uv6yN2JmhocevwUf" /><figcaption>Photo by <a href="https://unsplash.com/@growtika?utm_source=medium&amp;utm_medium=referral">Growtika</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Now we’ll see where does deep learning fit into the whole landscape of generative AI. Generative AI uses the power of deep learning to create new content, spanning text, images, audio, video, and beyond. Deep learning techniques, particularly those centered around neural networks, are the engine behind these generative AI models.</p><p>In our previous articles, we discussed about foundation models and the large language models, which is nothing but one of the exciting type of foundation model. Now it is the foundation model which uses deep learning. These models are trained on massive datasets that allow them to learn complex patterns and perform a variety of tasks across different domains. They’re incredibly powerful machine learning models, trained on a massive scale, often using vast amounts of unlabeled data. This training allows them to develop a broader understanding of the world, capturing intricate patterns and relationships within the data they consume. In layman terms, foundation model is a top-ranker in a class, like a student who has read everything in the entire library, absorbing knowledge from all the books, articles, websites. This deep and extensive learning allows foundation models to be adapted to a wide range of tasks.</p><p>To summarize, deep learning provides the core technology, foundation models are the powerful architectures built on deep learning, and generative AI is the application of these models to create new and original content.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0438bdeab5fd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI data engineering lifecycle from Google Cloud lens]]></title>
            <link>https://medium.com/@learnwithmohitvyas/ai-data-engineering-lifecycle-from-google-cloud-lens-ca7988012434?source=rss-5811f2e1bf2c------2</link>
            <guid isPermaLink="false">https://medium.com/p/ca7988012434</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[google-cloud-platform]]></category>
            <category><![CDATA[data-engineering]]></category>
            <dc:creator><![CDATA[Learn With Mohit Vyas]]></dc:creator>
            <pubDate>Sun, 08 Mar 2026 14:21:02 GMT</pubDate>
            <atom:updated>2026-03-08T14:21:02.762Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>AI data engineering lifecycle from Google Cloud lens</strong></p><p>Today we will be learning about AI data engineering lifecycle from the lens of Google Cloud. So what are the different steps when you want to design a system which can provide the data for your AI use cases? So we’ll learn about that.</p><figure><img alt="Bike Toy" src="https://cdn-images-1.medium.com/max/1024/0*ZTfv_Cta4wyxPYf9" /><figcaption>Photo by <a href="https://unsplash.com/@coopery?utm_source=medium&amp;utm_medium=referral">Mohamed Nohassi</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>The first step, gather your data. Data gathering, also known as data ingestion, involves collecting raw data from various sources. You want to collect the data based on what kind of outcome you want to achieve from the data. Different Google Cloud tools which are available for data ingestion are Pub/Sub, which can handle the live streaming of data regardless of the structure of data, Cloud Storage, which is well suited for unstructured data, and Cloud SQL and Cloud Spanner, which are used to manage the structured data.</p><p>The second step, prepare your data. Data preparation is the process of cleaning and transforming raw data into a usable format for analysis or model training. This involves formatting the data and labeling it properly. Different Google Cloud tools which are available for preparing of data are BigQuery, which can be used for data analysis, and Data Catalog, which can be used for data governance. These tools help prepare the data for ML models. Using BigQuery, we can filter data, correct its inconsistencies, and handle missing values. With the Data Catalog, we can find relevant data for the ML projects. It’s a centralized repository to easily discover data sets in your organization.</p><p>Third step, train your model. So the process of creating your ML model using data is called model training. Google Cloud provides Vertex AI platform, which provides a managed environment for training ML models. Using Vertex AI, we can set parameters for building our models using pre-built containers for popular machine learning frameworks, custom training jobs, and tools for model evaluation. It also provides access to powerful computing resources to make the model training process faster.</p><p>The fourth step, deploy and predict. So model deployment is the process of making a trained model available for use. So Google Cloud tool which can be used for this step is Vertex AI again. It simplifies the whole process by providing tools to put the model into action for generating predictions. It includes scaling the deployment, which means adjusting the resources allocated to the model to handle varying amounts of usage.</p><p>Fifth step, manage your model. Model management is the process of managing and maintaining your models over time. So Google Cloud tools which are used for the entire lifecycle of ML models are Versioning, to keep track of different versions of the model. Performance tracking, to review the model metrics to check the model’s performance. Drift monitoring, to watch for changes in the model’s accuracy over time. Data management, use Vertex AI features to manage the data features the model uses. Storage, use Vertex AI model garden to store and organize the models in one place. Automate, use Vertex AI pipelines to automate your machine learning tasks.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ca7988012434" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Dynamic DAG generation in Apache Airflow]]></title>
            <link>https://medium.com/@learnwithmohitvyas/dynamic-dag-generation-in-apache-airflow-e9b653058860?source=rss-5811f2e1bf2c------2</link>
            <guid isPermaLink="false">https://medium.com/p/e9b653058860</guid>
            <category><![CDATA[google-cloud-platform]]></category>
            <category><![CDATA[directed-acyclic-graph]]></category>
            <category><![CDATA[metadata]]></category>
            <category><![CDATA[airflow]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Learn With Mohit Vyas]]></dc:creator>
            <pubDate>Fri, 13 Feb 2026 17:08:51 GMT</pubDate>
            <atom:updated>2026-02-13T17:08:51.435Z</atom:updated>
            <content:encoded><![CDATA[<p>Today we will talk about dynamic DAG generation in Apache Airflow. So I came across a requirement where we had to create n number of DAGs in Airflow, but in an automated format. So we devised a technique driven by metadata, which we will see now.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*mwJ1fiULk4Cuub8V" /><figcaption>Photo by <a href="https://unsplash.com/@markuswinkler?utm_source=medium&amp;utm_medium=referral">Markus Winkler</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>The target is to create DAGs dynamically based on the data which is available. For dynamic DAG creation on the fly in Airflow we need first, metadata for the DAG which is to be created, second, a template in the form of a Jinja, which can take and render the output DAG file, which can then create the DAGs on Airflow UI.</p><figure><img alt="Web development" src="https://cdn-images-1.medium.com/max/1024/0*oy_omog1cZiNSjMh" /><figcaption>Photo by <a href="https://unsplash.com/@juanjodev02?utm_source=medium&amp;utm_medium=referral">Juanjo Jaramillo</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Now first we will talk about the metadata which is in the The metadata which is used for the DAG will contain the details which are required to create the DAG. It will have the DAG name, the details of the tasks which are to be created, the schedule at which the DAG is to be triggered, and also the details of the file, suppose it is a SQL file which is to be triggered in that particular task. So you can have parent task and the child tasks which can be linked together. In order to form the linkage, you have to define the dependency in the metadata table. You can have the parent task, which will be defined as with the keyword as parent, and then a child tasks, series of child tasks, which will have a reference of their parent task name.</p><figure><img alt="Close-up of steel chain link on white background." src="https://cdn-images-1.medium.com/max/1024/0*a4GooDsRK7wc2w-S" /><figcaption>Photo by <a href="https://unsplash.com/@edge2edgemedia?utm_source=medium&amp;utm_medium=referral">Edge2Edge Media</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Now once you define the metadata in the form of a table, the same metadata can be utilized and be mapped with the template, which we’ll be talking about now. So template will be in the form of a Jinja file or the Jinja template, which can have the predefined sections along with the placeholders. With the data coming from the metadata table, once the data is mapped from the metadata table to the Jinja template, it will be rendered in the form of the dag.py file with the name of the parent task, which we defined in the metadata table.</p><p>Once the DAG.py file is created, it can be automatically placed at the storage location from where the Airflow DAG is created. Now, assuming that your DAG is created, but you want to make some changes in the DAG, so you want to edit the properties or basically the metadata values which helped you create the DAG. So what will you do? You would need to change the metadata values in the table and then again re-generate or refresh the DAG using the Jinja template. So for that, we can have operational columns in the metadata table, like the DAG creation timestamp and dag last updated timestamp. So once you create the DAG for the first time, the created and the last updated timestamp should be same, but once you start editing and refreshing the DAG and editing the metadata, updating the entry for that particular task, the creation time will remain the same as the original creation time, but the last update timestamp will keep on updating based on the last edit in the metadata table.</p><p>In order to automate this entire process, you’ll have to create a DAG, which we’ll call an automated DAG creation job, which will keep on reading the table of any new entry into the table or in case of any existing entries in the table. If it detects change in the metadata table, it will fetch that change in the table and trigger the Jinja template, map the values, and generate the output file in the form of DAG.py file for the respective change which has happened in that table.</p><p>So to conclude, with this approach, you can create any number of DAGs in Apache Airflow without actually having to code the DAGs manually, but by just making an entry in the form of metadata in the metadata table, which will be rendered in the form of DAG.py file.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e9b653058860" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Document creation using Python]]></title>
            <link>https://medium.com/@learnwithmohitvyas/document-creation-using-python-9bdbd33a5c27?source=rss-5811f2e1bf2c------2</link>
            <guid isPermaLink="false">https://medium.com/p/9bdbd33a5c27</guid>
            <category><![CDATA[python-docx]]></category>
            <category><![CDATA[automation]]></category>
            <category><![CDATA[reporting]]></category>
            <category><![CDATA[docxtpl]]></category>
            <category><![CDATA[python]]></category>
            <dc:creator><![CDATA[Learn With Mohit Vyas]]></dc:creator>
            <pubDate>Sat, 07 Feb 2026 16:25:02 GMT</pubDate>
            <atom:updated>2026-02-07T16:25:02.917Z</atom:updated>
            <content:encoded><![CDATA[<p>Today we will be talking about how to create documents using Python.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*sBeenqyQeEpeh5Jo" /><figcaption>Photo by <a href="https://unsplash.com/@bk71?utm_source=medium&amp;utm_medium=referral">Bernd Klutsch</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Recently, I came across a requirement where we had to create reporting solution using Python. We explored many options like inbuilt tools and the other external vendor tools, but they were coming with a cost. So what we finalized is what we’ll see now.</p><p>Let’s assume you have data coming from different sources and you are curating the data into a layer which you call as the data lake or the cloud data warehouse. And then from that warehouse or the data lake, you have to create documents or reports or any such thing which can be viewed by the leadership. So what will you do in that case? How will you automate the entire process? That is what we’ll be talking about in this article.</p><p>In order to create any document, report, or any such thing from the data which is present in your cloud data warehouse or data lake, you’ll have to first know what kind of document to create or the structure of the document to be rendered or created which is called a template.</p><figure><img alt="Feuille de notation produit cannabis et CBD customisable avec votre logo ou vos informations" src="https://cdn-images-1.medium.com/max/1024/0*4ANg7D6DgSuFcQvM" /><figcaption>Photo by <a href="https://unsplash.com/@testeurdecbd?utm_source=medium&amp;utm_medium=referral">Testeur de CBD</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Once you have the template or the structure of the document which is to be created, you would need something as the intermediate copy of the data, which can be in any form or format, let it be XML or JSON. Now you have the template ready and also the data which is to be rendered in the template. What next? The next is the mapping of the data present in the JSON or the XML with the template.</p><figure><img alt="Code example of JSON file" src="https://cdn-images-1.medium.com/max/1024/0*_weafSMKknAo-A6F" /><figcaption>Photo by <a href="https://unsplash.com/@flowforfrank?utm_source=medium&amp;utm_medium=referral">Ferenc Almasi</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Once you know which data is to be placed at which position in the template or the final document, you’ll be having the placeholder in place in the template. Each placeholder will be mapped one-to-one with the data which is present in your XML or JSON file. Once all the elements of the JSON or the XML file are mapped with the template in the form of placeholders, you are ready to finalize your template.</p><p>In order to render the data present in XML or JSON onto the template to create the final document, we’ll be using native Python libraries. One such library, which is natively available in Python is Docxtpl. You can use this library and its functions in order to generate the final target document from the data which is available in your JSON or XML.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Y1OCITlTQ-0Zq9hc" /><figcaption>Photo by <a href="https://unsplash.com/@rubaitulazad?utm_source=medium&amp;utm_medium=referral">Rubaitul Azad</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9bdbd33a5c27" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to use foundation models for Generative AI?]]></title>
            <link>https://medium.com/@learnwithmohitvyas/how-to-use-foundation-models-for-generative-ai-03deff14d164?source=rss-5811f2e1bf2c------2</link>
            <guid isPermaLink="false">https://medium.com/p/03deff14d164</guid>
            <category><![CDATA[prompt-engineering]]></category>
            <category><![CDATA[writing-prompts]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[genai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Learn With Mohit Vyas]]></dc:creator>
            <pubDate>Wed, 28 Jan 2026 03:54:42 GMT</pubDate>
            <atom:updated>2026-01-28T03:54:42.664Z</atom:updated>
            <content:encoded><![CDATA[<p>As we discussed in our previous article, <em>“How Generative AI actually works?”,</em> foundation models are trained on vast amount of data, but how do we make use of that knowledge that the foundation model has? The answer lies in prompting.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Yn_vrRM_7SwuoDUZ" /><figcaption>Photo by <a href="https://unsplash.com/@solenfeyissa?utm_source=medium&amp;utm_medium=referral">Solen Feyissa</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Input given to a foundation model to produce an output is called a prompt. While all the AI models have inputs or prompts, foundation models are much more flexible in what you can prompt.</p><p>For example, some AI models can have a strict requirement of the type of inputs that you can feed in. Like only the email IDs or the images or some can handle only numbers. But this is not the case for foundation models because they can handle a varied variety of data, especially the multimodal foundation models.</p><p>Let’s talk about some sample prompts that you can give to the Foundation model. Example, if you ask a question, the output is the answer to the question. If you request an image based on the description, then the output is the generated image. If you request a summary of a file, then the output would be the file summary. If you request to debug a code with errors, then the output would be the corrected code without errors.</p><p>You must have heard of the word called prompt engineering or prompt design. What is it? So prompt engineering is the combination of different techniques and the methodologies which are used to write better prompts. Creating these prompts for Gen AI models is an art and a science in itself. Prompting is one of the most important skills in maximizing the value of the generative AI models and the tools.</p><p><em>To conclude, we can say that foundation models are very powerful tools which can be accessed through prompting, which is nothing but asking. You can unlock the potential of these models to do things like generating text or translating languages, write different kinds of creative content, and answering your questions in an informative way.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=03deff14d164" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How Generative AI actually works?]]></title>
            <link>https://medium.com/@learnwithmohitvyas/how-generative-ai-actually-works-0e225e618538?source=rss-5811f2e1bf2c------2</link>
            <guid isPermaLink="false">https://medium.com/p/0e225e618538</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[google-cloud-platform]]></category>
            <category><![CDATA[foundation-models]]></category>
            <category><![CDATA[genai]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Learn With Mohit Vyas]]></dc:creator>
            <pubDate>Tue, 27 Jan 2026 03:39:38 GMT</pubDate>
            <atom:updated>2026-01-27T03:39:38.934Z</atom:updated>
            <content:encoded><![CDATA[<p>Generative AI actually works on the basis of a concept which is called foundation models. Foundation models are large AI models which are trained on enormous datasets. This allows them to develop a deep understanding of the data and be adaptable to a wide range of downstream tasks.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*cSfwxEooW1ea4p89" /><figcaption>Photo by <a href="https://unsplash.com/@growtika?utm_source=medium&amp;utm_medium=referral">Growtika</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Understanding foundation models is critical because it represents the backbone of generative AI. They offer the potential to streamline processes, automate tasks, and unlock new possibilities across various business functions.</p><p>Traditionally, the AI models were trained for single purpose using specific data like filtering spam emails from the inbox. But foundation models are different. They are trained on massive amounts of data, which is diverse in nature, like text, images, and code, which enables them to adapt to many different tasks. This kind of flexibility is what makes them so groundbreaking.</p><p>The most famous type of foundation model is probably the LLMs or the large language models. They are specialized type of foundation models that focus specifically on language.</p><p>Now let’s talk about the key features of foundation models.</p><p>The first, trained on diverse data. Foundation models are trained on a wide variety of data allowing them to learn general patterns and relationships that can be applied to different tasks.</p><p>Second, flexible. With foundation models, one AI model can support a wide range of use cases.</p><p>Third, adaptable. Foundation models can be specialized for specific and particular domains or use cases through additional targeted learning or training.</p><p>Some examples of foundation models include Gemini, Imagen, and Chirp.</p><p>To conclude, foundation models are large-scale, general-purpose models which are trained on massive amount of data. They can be adapted to a wide range of downstream tasks. In contrast, traditional machine learning models are often designed for specific tasks and trained on smaller, more focused datasets.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0e225e618538" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI image generation - Google Cloud Imagen]]></title>
            <link>https://medium.com/@learnwithmohitvyas/ai-image-generation-google-cloud-imagen-2bfa9f30d9d0?source=rss-5811f2e1bf2c------2</link>
            <guid isPermaLink="false">https://medium.com/p/2bfa9f30d9d0</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[photography]]></category>
            <category><![CDATA[ai-image-generation]]></category>
            <dc:creator><![CDATA[Learn With Mohit Vyas]]></dc:creator>
            <pubDate>Sun, 25 Jan 2026 03:00:22 GMT</pubDate>
            <atom:updated>2026-01-25T06:28:21.064Z</atom:updated>
            <content:encoded><![CDATA[<p>Google Cloud Imagen is a very powerful image generation tool that allows businesses to create high-quality, customized, and localized product imagery at scale.</p><figure><img alt="YouTube.com/@DreySantesson" src="https://cdn-images-1.medium.com/max/1024/0*VW8lh4owDgpRLRtV" /><figcaption>Photo by <a href="https://unsplash.com/@santesson89?utm_source=medium&amp;utm_medium=referral">Andrea De Santis</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Imagin is currently being used by many big brands to customize their product photos on their websites to save time and ensure that the photos are locally relevant to their customers.</p><p>One such example is a global sports brand which partnered with Google Cloud to revolutionize its e-commerce product imagery. It can now create high-quality images of their products in different settings, backgrounds, colors, and styles without the need of expensive photo shoots for the products.</p><p>This approach not only saves time and resources but also allows for greater flexibility and creativity in showcasing their products online on their website. The result is a more engaging customer experience for its customers and it drives sales and the brand loyalty.</p><p>For example, a new product is launched on their website, so they will provide a sample image of the product to Imagen and give a prompt about what kind of AI backgrounds they want, and then they click on Generate AI background for that product image. Now the same product the company wants to launch in different markets across different regions. So they will adjust the prompt accordingly and the imagen will generate the product image with the AI background relevant to that particular market and region.</p><figure><img alt="Spending money on things we might not need." src="https://cdn-images-1.medium.com/max/1024/0*f1cSEPNWr8iZqKYU" /><figcaption>Photo by <a href="https://unsplash.com/@charlesdeluvio?utm_source=medium&amp;utm_medium=referral">charlesdeluvio</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>The result is, the company is already seeing the benefits of this strategy by having a 10% increase in their click-through rate. So imagen is cost-effective, creates high-quality and localized visuals at scale, and it’s a perfect choice for them.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ew5GU8Kep5Oy0SDG" /><figcaption>Photo by <a href="https://unsplash.com/@towfiqu999999?utm_source=medium&amp;utm_medium=referral">Towfiqu barbhuiya</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2bfa9f30d9d0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Importance of Generative AI]]></title>
            <link>https://medium.com/@learnwithmohitvyas/importance-of-genai-909e1ca8f5d5?source=rss-5811f2e1bf2c------2</link>
            <guid isPermaLink="false">https://medium.com/p/909e1ca8f5d5</guid>
            <category><![CDATA[genai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[generative-ai-use-cases]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[productivity]]></category>
            <dc:creator><![CDATA[Learn With Mohit Vyas]]></dc:creator>
            <pubDate>Fri, 23 Jan 2026 05:15:35 GMT</pubDate>
            <atom:updated>2026-01-23T05:26:01.546Z</atom:updated>
            <content:encoded><![CDATA[<p>Today we’ll be talking about the importance of using generative AI tools and technologies in today’s era of companies.</p><figure><img alt="3D render of AI and GPU processors" src="https://cdn-images-1.medium.com/max/1024/0*kuT1Enr5PwAGBMzs" /><figcaption>Photo by <a href="https://unsplash.com/@omilaev?utm_source=medium&amp;utm_medium=referral">Igor Omilaev</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>So, the companies or the organizations which are not utilizing the generative AI tools and technologies might risk lagging behind as compared to their peers. That is what is currently being observed in the current market scenario. So the companies which educate their employees to utilize the generative AI tools and technologies are accelerating and boosting their capabilities to 100x the companies which are not utilizing those tools and technologies.</p><p>So now we’ll be talking about the things where generative AI can be used.</p><p>So the first thing that comes to mind is the document translation. So if a particular document is in a particular language and you want to get it translated to, suppose, from English to French, so that can be one of the use cases which you can use generative AI for.</p><p>The second one would be, similarly on the document translation, would be the document summarization. So you provide a document to generative AI and it will get you a summary of the particular document. And when the same is being done by human mind, that might miss out on some of the aspects of the points which are mentioned in the document.</p><p>And the third one would be, from a document or from a particular piece of conversation, the generative AI can generate a video for you. It’s like a podcast where two people are talking about. So that is also one of the use cases.</p><p>And the fourth use case, which is a major one, is the code building for the developers. So we can 100x the productivity which a normal person would not be able to do without using generative AI tools and technologies. So building codes become 100 times faster and you can boost your productivity and your importance in the organization.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=909e1ca8f5d5" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>