<?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 Tina Kanth on Medium]]></title>
        <description><![CDATA[Stories by Tina Kanth on Medium]]></description>
        <link>https://medium.com/@tina.snpraveen?source=rss-40585dfb0027------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*g5J7RrA-dylOUla_OseMAQ.jpeg</url>
            <title>Stories by Tina Kanth on Medium</title>
            <link>https://medium.com/@tina.snpraveen?source=rss-40585dfb0027------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 14:01:05 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@tina.snpraveen/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[Prompt Power: The New Skill Every Productive Developer Needs]]></title>
            <link>https://medium.com/@tina.snpraveen/prompt-power-the-new-skill-every-productive-developer-needs-14a2ae74ce22?source=rss-40585dfb0027------2</link>
            <guid isPermaLink="false">https://medium.com/p/14a2ae74ce22</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[prompt]]></category>
            <category><![CDATA[vibe-coding]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ai-agent]]></category>
            <dc:creator><![CDATA[Tina Kanth]]></dc:creator>
            <pubDate>Fri, 29 Aug 2025 16:45:07 GMT</pubDate>
            <atom:updated>2025-08-29T16:45:07.715Z</atom:updated>
            <content:encoded><![CDATA[<p>In today’s era of AI-assisted development, mastering the art of prompting your coding agent can transform it from a helpful sidekick into a productivity powerhouse. As a software engineer who’s tested everything from GitHub Copilot to custom in-house bots, I’ve distilled the most effective, everyday prompting strategies that consistently deliver cleaner code, faster feature builds, and smarter debugging. Ready to multiply your output and reclaim hours in your sprint?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JcCn3IMibXvhPH8bVPnqNA.png" /></figure><h3>1. Establish a Stable Prompt Framework</h3><p>A prompt template should include core instructions that never change. This static layer ensures the AI always follows your primary guidelines, whether it’s maintaining tone, focusing on a topic, or adhering to a workflow.</p><ul><li>Define the overall goal: explain the assistant’s purpose in a single sentence.</li><li>Lock down formatting rules: specify how lists, code, or citations should appear.</li><li>Reinforce any non-negotiable constraints: word limits, prohibited content, style guidelines.</li></ul><p>By keeping these elements fixed, you build a reliable foundation for every interaction.</p><h3>2. Inject Dynamic Placeholders with Double Brackets</h3><p>Once your static framework is in place, add placeholders for the details that change from one prompt to the next. Use double-bracket notation for clarity:</p><ul><li>{{user_input}} for the current question or request</li><li>{{user_history}} for previous interactions or context</li><li>{{system_generated_data}} for facts, stats, or dates provided at runtime</li></ul><p>This pattern makes your template both flexible and easy to maintain. Updating a single placeholder automatically tailors the prompt to new situations without rewriting the entire prompt.</p><h3>3. Assign a Persona to Guide the Conversation</h3><p>Specifying a persona steers the AI’s perspective and language style. Tailor the role to your needs:</p><ul><li>“You are a Senior C# developer” to focus on code review and best practices</li><li>“You are a Financial Analyst” to analyze balance sheets and market trends</li><li>“You are a Customer Support Specialist” to handle complaints with empathy</li></ul><p>A clear persona directive helps the AI choose relevant vocabulary, tone, and domain knowledge, resulting in more on-target responses.</p><h3>4. Use Multishot Prompting with Embedded Examples</h3><p>When you need consistent formatting or behavior, show rather than tell. Include one or more annotated examples in your prompt:</p><p>&lt;example&gt;</p><p>User: Generate an email summarizing Q2 sales.</p><p>Assistant:</p><ul><li>Subject: Q2 Sales Highlights</li><li>Body: In Q2, revenue increased by 12%… &lt;/example&gt;</li></ul><p>Embedding examples as &lt;example&gt; blocks gives the AI concrete patterns to emulate. This multi shot technique reduces ambiguity and guides the model toward your desired style.</p><h3>5. Invoke Chain-Of-Thought for Complex Reasoning</h3><p>For intricate problems, prompt the AI to reveal its reasoning. Preface your prompt with a cue like “With step-by-step thinking,” and you’ll get a transparent breakdown of logic:</p><p>With step-by-step thinking, explain how you would approach optimizing a database schema.</p><p>This approach not only provides deeper insight but often yields more accurate and trustworthy results for multifaceted tasks.</p><h3>6. Define Success Criteria in Structured Formats</h3><p>Closing the loop with explicit output requirements can dramatically improve consistency. Specify the exact structure you expect — JSON, XML, or a custom schema:</p><p>Provide the response in JSON with fields:</p><pre>{<br>  &quot;summary&quot;: &quot;...&quot;,<br>  &quot;key_metrics&quot;: [...],<br>  &quot;recommendations&quot;: [...]<br>}</pre><p>By setting success criteria, you give the AI a clear checklist, making post-processing simpler and reducing the need for manual corrections.</p><p>Start your next project by sketching out this framework. You’ll find that a little upfront planning pays off in faster iteration cycles, fewer misunderstandings, and ultimately, an AI assistant that feels custom-built for your needs.</p><p>Ready to level up your prompt game? Begin by drafting your core template today and watch your AI interactions become sharper, more reliable, and infinitely more useful.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=14a2ae74ce22" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The AI Paradigm Shift: Moving Beyond Traditional ML to Generative AI]]></title>
            <link>https://medium.com/@tina.snpraveen/the-ai-paradigm-shift-moving-beyond-traditional-ml-to-generative-ai-de001e2f1416?source=rss-40585dfb0027------2</link>
            <guid isPermaLink="false">https://medium.com/p/de001e2f1416</guid>
            <category><![CDATA[generative-ai-use-cases]]></category>
            <category><![CDATA[ai-tech]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[innovation]]></category>
            <category><![CDATA[gen-ai-chatbots]]></category>
            <dc:creator><![CDATA[Tina Kanth]]></dc:creator>
            <pubDate>Tue, 15 Jul 2025 10:54:56 GMT</pubDate>
            <atom:updated>2025-07-15T10:54:56.607Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PDQfqkPx-E2O8D6IdyBrQg.jpeg" /></figure><p>Artificial Intelligence (AI) has long been a transformative force, quietly powering systems that analyze data, automate tasks, and support decision-making. Traditional AI, often referred to as rule-based, relies on predefined logic and human-crafted rules to mimic cognitive functions such as learning and problem-solving. These systems are designed to interpret inputs, analyze patterns, and optimize outcomes based on structured data and explicit goals.</p><p>However, the shift from deterministic, logic-driven systems to probabilistic, generative models marks a pivotal evolution in the AI landscape. Unlike traditional AI, generative AI is capable of creating entirely new content by leveraging advanced architectures like transformers and diffusion models to understand context, generate coherent outputs, and adapt across domains, making it a cornerstone of the next generation of intelligent systems</p><h3>Fundamental Building Blocks of a Traditional AI</h3><ol><li>Input Layer — Receives raw input data and passes it to the network.</li><li>Weights and Biases — Learnable parameters that define the strength and shift of neuron connections.</li><li>Hidden Layers — Intermediate layers that transform data using learned weights and activation functions.</li><li>Activation Functions — Introduce non-linearity to help the network learn complex patterns.</li><li>Output Layer — Produces the final prediction or classification result.</li><li>Loss Function — Measures the error between predicted and actual outputs.</li><li>Optimizer — Updates weights and biases to minimize the loss function.</li><li>Backpropagation — Computes gradients to guide the optimizer in updating model parameters.</li></ol><h3>Fundamental Building Blocks of Generative AI</h3><ol><li>Foundational Models — Large-scale pre-trained models that serve as the base for a wide range of downstream tasks.</li><li>Frameworks — Software libraries and tools (e.g., PyTorch, TensorFlow) that support model development, training, and deployment.</li><li>Fine-Tuning — The process of adapting a pre-trained model to a specific task or domain using additional labeled data.</li><li>Embeddings and Labeling — Techniques for representing data in a vector space and annotating it for supervised learning.</li><li>Synthetic Data — Artificially generated data used to augment training datasets or simulate rare scenarios.</li><li>Prompt Engineering — Crafting effective inputs to guide the behavior and output of generative models.</li><li>Agents &amp; Automation — Systems that use AI to autonomously perform tasks, make decisions, or interact with environments.</li></ol><h3>Use Cases for Traditional AI</h3><ol><li>Anomaly Detection — Identifies unusual patterns or outliers in data for fraud detection, system monitoring, etc.</li><li>Predictive Modeling — Forecasts future outcomes based on historical data using statistical and machine learning techniques.</li><li>Personalized Recommendations — Suggests products, content, or actions based on user behavior and preferences.</li></ol><h3>Use Cases for Generative AI</h3><ol><li>Content Creation (Text, Video/Image Generation) — Automatically generates human-like text, realistic images, videos, and other creative content.</li><li>Virtual Assistants — Powers conversational agents that understand and respond to natural language in real time.</li><li>Code Agents — Generate, explain, or debug code to assist developers in software engineering tasks.</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=de001e2f1416" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Streamlining Data Workflows and Minimizing Big Data Costs with Delta Lake CDF]]></title>
            <link>https://medium.com/@tina.snpraveen/streamlining-data-workflows-and-minimizing-big-data-costs-with-delta-lake-cdf-c46bf47b15df?source=rss-40585dfb0027------2</link>
            <guid isPermaLink="false">https://medium.com/p/c46bf47b15df</guid>
            <category><![CDATA[big-data]]></category>
            <category><![CDATA[big-data-analytics]]></category>
            <category><![CDATA[architectural-design]]></category>
            <category><![CDATA[azure]]></category>
            <category><![CDATA[delta-lake]]></category>
            <dc:creator><![CDATA[Tina Kanth]]></dc:creator>
            <pubDate>Fri, 11 Jul 2025 01:46:18 GMT</pubDate>
            <atom:updated>2025-07-11T01:46:18.228Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/313/1*SpGlYdkSJi16LsnwJKB3uQ.png" /></figure><p>Delta Lake extends the capabilities of traditional Parquet-based data lakes by introducing ACID transactions, schema enforcement, and evolution, and time travel for historical data access. These features make it a robust and reliable solution for managing large-scale data pipelines. Delta tables are optimized for both streaming and batch ingestion workflows, making them ideal for modern data lake architectures.</p><p>The Delta Lake Change Data Feed (CDF) enables efficient tracking of row-level changes — inserts, updates, and deletes — within Delta tables. This feature is designed to be high-performance, scalable, and reliable, making it well-suited for modern data engineering workflows.</p><p>CDF is particularly valuable in scenarios such as auditing, data quality validation, debugging, and incremental data processing. Instead of recomputing entire datasets, downstream systems can consume only the changed records, enabling minimal updates to aggregations and derived tables. This significantly reduces compute costs and improves pipeline efficiency.</p><ul><li>Atomic Availability of Change Data:<br>Change data is committed as part of the same transaction that writes to the Delta table. This ensures that change records become available simultaneously with the new or updated data, maintaining strong consistency guarantees.</li><li>Minimal Performance Overhead:<br>The generation of change data occurs inline during query execution, introducing no significant performance impact. Since change records are typically much smaller than the full set of rewritten files, the additional storage and compute overhead is minimal.</li><li>Retention and Cleanup:<br>Change data adheres to the same retention policies as historical table versions. As such, it is subject to cleanup via the VACUUM operation once it falls outside the configured retention window, ensuring efficient storage management over time.</li></ul><p>One of the most impactful use cases of Delta Lake Change Data Feed (CDF) is in high-throughput Spark Streaming pipelines that process millions of events per hour. In many real-world scenarios, only a subset of the ingested data is relevant for downstream processing. Without CDF, identifying these changes requires scanning the entire Delta table repeatedly to detect new inserts, updates, or deletes — an approach that is both resource-intensive and costly.</p><p>With CDF enabled, this overhead is drastically reduced. Instead of scanning the full dataset, CDF provides a stream of row-level changes, allowing downstream consumers to process only the relevant deltas. This optimization leads to significant savings in compute time, storage I/O, and overall infrastructure cost.</p><p>A particularly compelling example is in near-real-time processing for systems with strict API limits, egress constraints, or connection quotas, such as those interacting with transactional platforms. Previously, the system had to read the entire Delta Lake, often hundreds of gigabytes, every hour to extract changes. With CDF, the change data is now delivered at a much smaller volume — just 2–5 GB per hour — eliminating the need for full table scans.</p><p>This “read-once” model not only improves performance but also enables scalable and cost-effective data engineering. The result is a dramatic reduction in latency and operational overhead, making CDF a critical feature for modern big data architectures.</p><p>Here is an example of architectural design for one of the use cases as above:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8hLSl0K6sON01aBV_8bZdw.png" /><figcaption>Architecture diagram of data workflow with CDF Enabled</figcaption></figure><p>These are some useful code snippets while working with CDF capabilities</p><ol><li>Enable the change data feed for a new delta table:</li></ol><pre>CREATE TABLE student (id INT, name STRING, age INT) TBLPROPERTIES (delta.enableChangeDataFeed = true)</pre><p>2. Enable change data feed for the existing table:</p><pre>ALTER TABLE myDeltaTable SET TBLPROPERTIES (delta.enableChangeDataFeed = true)</pre><p>3. Read changes in the batch queries:(Scala)</p><pre>// version as ints or longs<br>spark.read.format(&quot;delta&quot;)<br>  .option(&quot;readChangeFeed&quot;, &quot;true&quot;)<br>  .option(&quot;startingVersion&quot;, 0)<br>  .option(&quot;endingVersion&quot;, 10)<br>  .table(&quot;myDeltaTable&quot;)<br><br>// timestamps as formatted timestamp<br>spark.read.format(&quot;delta&quot;)<br>  .option(&quot;readChangeFeed&quot;, &quot;true&quot;)<br>  .option(&quot;startingTimestamp&quot;, &quot;2021-04-21 05:45:46&quot;)<br>  .option(&quot;endingTimestamp&quot;, &quot;2021-05-21 12:00:00&quot;)<br>  .table(&quot;myDeltaTable&quot;)<br><br>// providing only the startingVersion/timestamp<br>spark.read.format(&quot;delta&quot;)<br>  .option(&quot;readChangeFeed&quot;, &quot;true&quot;)<br>  .option(&quot;startingVersion&quot;, 0)<br>  .table(&quot;myDeltaTable&quot;)<br><br>// path based tables<br>spark.read.format(&quot;delta&quot;)<br>  .option(&quot;readChangeFeed&quot;, &quot;true&quot;)<br>  .option(&quot;startingTimestamp&quot;, &quot;2021-04-21 05:45:46&quot;)<br>  .load(&quot;pathToMyDeltaTable&quot;)</pre><p>4.Read changes in streaming queries (Scala)</p><pre>// providing a starting version<br>spark.readStream.format(&quot;delta&quot;)<br>  .option(&quot;readChangeFeed&quot;, &quot;true&quot;)<br>  .option(&quot;startingVersion&quot;, 0)<br>  .table(&quot;myDeltaTable&quot;)<br><br>// providing a starting timestamp<br>spark.readStream.format(&quot;delta&quot;)<br>  .option(&quot;readChangeFeed&quot;, &quot;true&quot;)<br>  .option(&quot;startingVersion&quot;, &quot;2021-04-21 05:35:43&quot;)<br>  .load(&quot;/pathToMyDeltaTable&quot;)<br><br>// not providing a starting version/timestamp will result in the latest snapshot being fetched first<br>spark.readStream.format(&quot;delta&quot;)<br>  .option(&quot;readChangeFeed&quot;, &quot;true&quot;)<br>  .table(&quot;myDeltaTable&quot;)</pre><p>Have fun with CDF!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c46bf47b15df" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>