<?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 PARESH JOSHI on Medium]]></title>
        <description><![CDATA[Stories by PARESH JOSHI on Medium]]></description>
        <link>https://medium.com/@pareshjoshij?source=rss-11cc0d1504f4------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*7nHQzD1zTaU1qg68</url>
            <title>Stories by PARESH JOSHI on Medium</title>
            <link>https://medium.com/@pareshjoshij?source=rss-11cc0d1504f4------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 09:25:42 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@pareshjoshij/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[Beyond the Hype: My Hands-On Journey with the Gemini API on Vertex AI]]></title>
            <link>https://medium.com/@pareshjoshij/beyond-the-hype-my-hands-on-journey-with-the-gemini-api-on-vertex-ai-7428601a71c5?source=rss-11cc0d1504f4------2</link>
            <guid isPermaLink="false">https://medium.com/p/7428601a71c5</guid>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[google-cloud-platform]]></category>
            <dc:creator><![CDATA[PARESH JOSHI]]></dc:creator>
            <pubDate>Tue, 05 Aug 2025 11:52:17 GMT</pubDate>
            <atom:updated>2025-08-05T11:52:17.491Z</atom:updated>
            <content:encoded><![CDATA[<p>The generative AI landscape is moving at lightning speed. Every week, a new model or tool promises to change everything. But as professionals, our value isn’t in watching the headlines — it’s in understanding the technology deeply enough to apply it effectively.</p><p>That’s why I dove headfirst into the <strong>“Explore Generative AI with the Gemini API in Vertex AI”</strong> path. My goal was simple: to move past the demos and get my hands dirty with the tools that businesses will rely on. This wasn’t just a course; it was a simulated client project, and the objective was to deliver real-world value.</p><p>Here are the three “game-changing” insights I’m taking away from this experience.</p><h4>1. Multimodality Isn’t a Gimmick — It’s a Business Solution</h4><p>We’ve all seen AI generate text or images. The real magic of Gemini is its native ability to reason across different types of information at once. Imagine an application for a logistics company: you upload a photo of a damaged package (image), add the tracking number (text), and ask the model to draft an apology email to the customer, estimate the item&#39;s value from the image, and simultaneously initiate a claim process (code/function call).</p><p>This isn’t a series of separate tasks; it’s one seamless, multimodal query. Learning to structure these prompts in Vertex AI felt like unlocking a new level of problem-solving.</p><h4>2. The Platform is as Important as the Model</h4><p>A brilliant model is useless if it lives in a sandbox. The true power of using Gemini within Vertex AI is the enterprise-grade infrastructure surrounding it.</p><ul><li><strong>Security &amp; Data Governance:</strong> You’re not sending your sensitive data into the void. Vertex AI provides the controls to build securely.</li><li><strong>Scalability:</strong> You can go from a single-user prototype to an application serving millions without re-architecting everything.</li><li><strong>Integration:</strong> It’s built to work with the entire Google Cloud ecosystem, from BigQuery to Cloud Storage.</li></ul><p>This is the crucial difference between a fun experiment and a deployable business asset.</p><h4>3. Prompt Engineering is the New Essential Skill</h4><p>Communicating with a model as sophisticated as Gemini is a nuanced art. It’s not just about asking a question; it’s about providing context, structuring examples (few-shot prompting), and clearly defining the desired output format.</p><p>My biggest takeaway was learning to “think” like the model, anticipating its ambiguities and guiding it toward the most valuable response. This skill is the bridge between the raw power of the AI and the specific, high-quality result a business needs.</p><p><strong>The Final Deal</strong></p><p>Completing this program was more than just earning a badge. It was about acquiring a strategic toolkit to design, build, and deploy the next generation of intelligent solutions. The hype around Generative AI is real, but the opportunity is in the application. And with tools like Gemini on Vertex AI, we’re finally equipped to seize it.</p><p><em>What are your thoughts on moving GenAI from experiment to enterprise? Share your experiences in the comments below!</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7428601a71c5" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Your Competition is Still Just Reading Words. We’re Reading the Whole Picture.]]></title>
            <link>https://medium.com/@pareshjoshij/your-competition-is-still-just-reading-words-were-reading-the-whole-picture-3d672a795b95?source=rss-11cc0d1504f4------2</link>
            <guid isPermaLink="false">https://medium.com/p/3d672a795b95</guid>
            <category><![CDATA[google-gemini]]></category>
            <category><![CDATA[google-cloud-platform]]></category>
            <category><![CDATA[students]]></category>
            <dc:creator><![CDATA[PARESH JOSHI]]></dc:creator>
            <pubDate>Mon, 04 Aug 2025 06:03:27 GMT</pubDate>
            <atom:updated>2025-08-04T06:03:27.663Z</atom:updated>
            <content:encoded><![CDATA[<p>In any high-stakes environment — a merger, a legal discovery, a market analysis — the winner is the one who sees the truth fastest. For years, we’ve been hamstrung by a critical intelligence gap. Our AI could read the words in a document, but the most crucial data — the story told in charts, the numbers buried in tables, the insights locked in diagrams — remained invisible to our machines.</p><p>This created a blind spot. A costly, high-risk blind spot.</p><p>That era is now over. Having just earned the <strong>“Inspect Rich Documents with Gemini Multimodality and Multimodal RAG”</strong> Skill Badge from Google Cloud, I’ve moved beyond simple text analysis into a new class of document intelligence.</p><h4>The Asset: True Contextual Understanding</h4><p>The core asset is <strong>multimodality</strong>. This isn’t just an upgrade; it’s a new sense. It’s the AI’s ability to see a document the way an expert human analyst does — holistically.</p><p>When we see a claim of “unprecedented growth,” we instinctively look to the accompanying graph for proof. Gemini now operates with that same instinct. It doesn’t just parse text; it interprets the visual evidence that gives the text its meaning. It reads the footnotes, analyzes the trend lines, and understands the data tables in a single, unified process.</p><h4>The Play: Verifiable, Defensible Insights with Multimodal RAG</h4><p>An asset is only valuable when deployed correctly. The strategic play here is <strong>Multimodal Retrieval-Augmented Generation (RAG)</strong>.</p><p>If standard RAG is like having a research assistant who can find text passages, Multimodal RAG is like having a senior analyst who can synthesize a complete intelligence briefing. It builds its answers by pulling evidence from every part of a document:</p><ul><li>A specific clause in the legal text.</li><li>A data point from a financial table.</li><li>The trajectory of a line in a performance chart.</li></ul><p>The result isn’t just an answer; it’s a conclusion backed by verifiable, multi-format evidence. During the hands-on labs, the real power became clear when Gemini could answer a complex query by citing both a paragraph on page 5 and a specific segment of a diagram on page 20. This is how you build defensible, data-driven decisions.</p><h4>The ROI: Where This Delivers a Decisive Edge</h4><p>This capability isn’t theoretical. It delivers a tangible return by closing the intelligence gap in critical operations:</p><ul><li><strong>Deal Diligence:</strong> Instead of spending weeks manually cross-referencing financial statements with narrative reports, you can get a synthesized risk assessment in minutes.</li><li><strong>Legal &amp; Compliance:</strong> Instantly verify that contractual claims are supported by the data in appendices and exhibits, drastically reducing review time and human error.</li><li><strong>Market Intelligence:</strong> Transform dense analyst reports and whitepapers into actionable summaries where insights from text and charts are already integrated.</li></ul><p>The training from Google Cloud has provided the tools to build these next-generation solutions. The competitive landscape is no longer about who can process the most words, but who can understand the complete picture. With Gemini, we’re now equipped to see it all.</p><p>#GenAIExchange #GoogleCloud #Gemini #GenerativeAI #Multimodality #RAG #AI #BusinessIntelligence #FutureOfWork #DataAnalysis</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3d672a795b95" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Passed a Google Cloud Challenge Lab. The Real Win Wasn’t the Badge.]]></title>
            <link>https://medium.com/@pareshjoshij/i-passed-a-google-cloud-challenge-lab-the-real-win-wasnt-the-badge-1579bcf61f80?source=rss-11cc0d1504f4------2</link>
            <guid isPermaLink="false">https://medium.com/p/1579bcf61f80</guid>
            <category><![CDATA[genai]]></category>
            <category><![CDATA[python]]></category>
            <category><![CDATA[streamlit]]></category>
            <category><![CDATA[vertex-ai]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <dc:creator><![CDATA[PARESH JOSHI]]></dc:creator>
            <pubDate>Sun, 03 Aug 2025 04:55:42 GMT</pubDate>
            <atom:updated>2025-08-03T04:55:42.193Z</atom:updated>
            <content:encoded><![CDATA[<p>There’s a unique thrill that comes with starting a challenge lab. The timer is ticking, the objectives are laid out, and you feel that surge of confidence: “I’ve got this.” My mission was to build and deploy a full-stack AI Chef application for the <strong>Develop GenAI Apps with Gemini and Streamlit</strong> challenge. I was flying through the initial tasks — configuring the API, writing the Python code, adding the UI elements. It was all going according to plan.</p><p>Then I hit the wall. And I hit it hard.</p><p>It was Task 3: “Test the application.” I ran the command streamlit run chef.py. The terminal told me everything was fine, giving me the local and network URLs. I clicked the Web Preview button, full of confidence.</p><p>And got this:</p><p>503 Service Unavailable. WebSocket connection failed.</p><p>My heart sank. The application, which should have been a beautiful AI-powered recipe generator, was a blank white page of failure. The timer, of course, kept ticking.</p><h3>The Anatomy of a Deceptive Bug</h3><p>This wasn’t a simple syntax error. Those are easy. The terminal would have screamed at me with a clear Traceback. No, this was something more sinister. This was a silent crash.</p><p>My debugging journey began. I followed the standard playbook:</p><ol><li><strong>Check the Obvious:</strong> Was my Project ID correct? Yes. Was the region right? Yes.</li><li><strong>Check the APIs:</strong> Maybe the Vertex AI API wasn’t enabled? I ran gcloud services list --enabled. Nope, it was active and waiting.</li><li><strong>Isolate the Code:</strong> This is where things got weird. I decided to bypass Streamlit and run the script directly with python3 chef.py. To my shock, it ran perfectly. The terminal printed &quot;Vertex AI initialized successfully&quot; and &quot;GenerativeModel loaded successfully.&quot;</li></ol><p>This was the ultimate developer paradox: <strong>My code was perfect, but it didn’t work.</strong></p><p>The script could connect to Google Cloud. It could load the Gemini model. But the moment I tried to serve it as a web app with Streamlit, it would die without a trace. The WebSocket error was just a symptom; my application was crashing so fast that the browser never even had a chance to connect.</p><h3>The “Aha!” Moment in a requirements.txt File</h3><p>After going in circles, I realized the problem wasn’t in the code I had written, but in the foundation it was built on — the project’s dependencies.</p><p>The initial requirements.txt file was missing a crucial library: google-cloud-aiplatform. My first attempt to fix this was to manually install it with pip. While this solved the ModuleNotFoundError, it created a new, invisible problem. By not specifying a version, pip had pulled the latest release, which was subtly incompatible with the older version of Streamlit used in the lab environment.</p><p>It was a classic case of version conflict, a silent killer that doesn’t show up as a code error but as a complete system failure.</p><p>The solution wasn’t to add more code, but to be more deliberate. I had to perform a “clean room” reset.</p><ol><li>I deleted the entire project folder.</li><li>I re-cloned the repository.</li><li>Crucially, I <strong>rewrote the </strong>requirements.txt<strong> file from scratch</strong>, specifying the exact, compatible versions of the libraries needed.</li><li>I created a fresh virtual environment and installed everything from this perfected requirements file.</li></ol><p>I held my breath, ran streamlit run chef.py one last time, and clicked the Web Preview link.</p><p>And there it was. My AI Chef application, alive and working perfectly.</p><h3>The Real Lesson</h3><p>I passed the lab and earned the badge, but my real takeaway wasn’t about Gemini or Streamlit. It was a powerful lesson in debugging. The bug forced me to look beyond my own code and analyze the entire ecosystem it lived in. It reminded me that a successful application isn’t just about correct logic; it’s about a stable, compatible, and predictable environment.</p><p>That struggle, that frustrating wall I hit, taught me more than a smooth run ever could have. It taught me persistence, methodical debugging, and the critical importance of managing dependencies.</p><p>And that’s a skill no course can teach you. You just have to live it.</p><p>It was a tough but rewarding challenge, and I’m proud to have the badge to show for it. You can view my verified credential on my Google Cloud Skills Boost profile:</p><p><a href="https://www.cloudskillsboost.google/public_profiles/108adecb-2579-45bd-83c9-2e07837cd585/badges/17275339"><strong>View My Verified Badge</strong></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1579bcf61f80" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[From Theory to Reality: My First Steps Building with Gemini and Imagen]]></title>
            <link>https://medium.com/@pareshjoshij/from-theory-to-reality-my-first-steps-building-with-gemini-and-imagen-52f66958c188?source=rss-11cc0d1504f4------2</link>
            <guid isPermaLink="false">https://medium.com/p/52f66958c188</guid>
            <dc:creator><![CDATA[PARESH JOSHI]]></dc:creator>
            <pubDate>Sat, 26 Jul 2025 04:38:37 GMT</pubDate>
            <atom:updated>2025-07-26T04:38:37.553Z</atom:updated>
            <content:encoded><![CDATA[<p>After learning the “secret language” of AI through prompt design, it was time to put my new communication skills to the test. The next stage of my quest was to move from theory to reality — to get my hands dirty and start building.</p><p>This led me straight to my second Google Cloud skill badge: <strong>“Build Real World AI Applications with Gemini and Imagen.”</strong> This wasn’t just about learning; it was about creating.</p><h3>Wielding the Power of AI Titans</h3><p>This course introduced me to two of the most impressive models in the Google ecosystem, and I got to see firsthand what they could do.</p><ul><li><strong>Gemini: The Multimodal Powerhouse</strong> What truly amazed me about Gemini is its ability to understand the world in a more human-like way. It’s not limited to just text. The course demonstrated how Gemini can seamlessly process and reason across text, images, and code all at once. This multimodal capability opens up a universe of possibilities for applications that can see, read, and understand complex information.</li><li><strong>Imagen: The AI Artist</strong> If Gemini is the brain, Imagen is the creative soul. This is the model that can take a simple text description and turn it into a vivid, high-quality image. The moment I wrote my first prompt — “a photorealistic image of an astronaut riding a horse on Mars” — and saw a unique, detailed image appear on my screen, everything clicked. This wasn’t just a tool; it was a creative partner.</li></ul><h3>The “Aha!” Moment</h3><p>The most valuable part of this course was its focus on “Real World Applications.” It bridged the gap between knowing <em>about</em> AI and knowing <em>how to use</em> AI. We walked through the practical steps of building solutions with these models, turning abstract concepts into tangible results.</p><p>The journey has been incredibly rewarding. It has solidified my passion for technology and my desire to build applications that are not only functional but also intuitive and maybe even a little bit magical.</p><p>My quest is far from over. I’m actively seeking opportunities to apply these new skills in a professional environment and contribute to building the next wave of AI-powered tools. If you’re working on interesting problems in this space, I would love to connect.</p><p>You can see more of my work and follow my journey on my portfolio at <strong>pareshjoshi.me</strong>.</p><p>Thanks for reading!</p><p><strong>#Gemini #Imagen #AI #GenerativeAI #GoogleCloud #Python #Developer #CareerChange #TechTransition #LearningInPublic #GenAIExchange</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=52f66958c188" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Art of Conversation: My Journey into Prompt Design with Vertex AI]]></title>
            <link>https://medium.com/@pareshjoshij/the-art-of-conversation-my-journey-into-prompt-design-with-vertex-ai-50d0b7d0a432?source=rss-11cc0d1504f4------2</link>
            <guid isPermaLink="false">https://medium.com/p/50d0b7d0a432</guid>
            <dc:creator><![CDATA[PARESH JOSHI]]></dc:creator>
            <pubDate>Sat, 26 Jul 2025 04:35:33 GMT</pubDate>
            <atom:updated>2025-07-26T04:35:33.734Z</atom:updated>
            <content:encoded><![CDATA[<p>As I navigate my career transition from business into the world of technology, I’ve learned that every great journey starts with a single, crucial step. For me, that first step into the vast landscape of Generative AI wasn’t about complex code or massive datasets; it was about learning how to communicate.</p><p>This led me to my first milestone: earning the <strong>“Prompt Design in Vertex AI”</strong> skill badge from Google Cloud. I discovered that “talking” to an AI is a skill in itself — an art form and a science.</p><h3>The Challenge: Moving Beyond Simple Questions</h3><p>Before this course, I thought you just asked an AI a question and it gave you an answer. I was wrong. To get truly powerful results, you can’t be a passive user; you have to be an active director. You have to become a <strong>Prompt Engineer</strong>.</p><p>The course was my training ground, and it was built on three powerful pillars:</p><ul><li><strong>Clarity is King:</strong> My biggest takeaway was that an AI model is a powerful engine, but it needs clear fuel. I learned to craft prompts that are specific, unambiguous, and provide all the necessary context. Instead of asking, “Tell me about cars,” I learned to ask, “Describe the key differences in engine performance between a 2023 Honda Civic and a 2023 Toyota Corolla.” The difference in output quality was astonishing.</li><li><strong>Structure is Power:</strong> The course introduced me to structured prompting techniques. The most impactful for me was “few-shot” prompting, where you provide the AI with a few examples of the input and desired output. This acts as a mini-training session, guiding the model to understand the exact format and style you need. It’s the difference between giving a chef a recipe versus just telling them to “make food.”</li><li><strong>It’s a Creative Process:</strong> While there’s a science to the structure, there’s a real art to finding the right words, tone, and persona to get the perfect result. It felt less like programming and more like being a creative director for a very intelligent, very literal assistant.</li></ul><h3>My Next Chapter</h3><p>Mastering the fundamentals of prompt design felt like learning the secret language of AI. It’s a foundational skill that I know will be invaluable as I continue my journey. It has given me the confidence to not just use AI, but to guide it effectively.</p><p>Now that I know how to talk to the models, my next step is to build with them.</p><p>If you’re also on a learning journey or are working on interesting problems in tech, I’d love to connect. You can follow my progress on my portfolio at <strong>pareshjoshi.me</strong>.</p><p>Thanks for reading!</p><p><strong>#PromptEngineering #VertexAI #GenerativeAI #AI #GoogleCloud #CareerChange #TechTransition #Developer #LearningInPublic #GenAIExchange</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=50d0b7d0a432" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>