<?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 Aravind C on Medium]]></title>
        <description><![CDATA[Stories by Aravind C on Medium]]></description>
        <link>https://medium.com/@aravindc42?source=rss-9dfee954f9db------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*LhXQsrA-Lg7ADW8z_Xd0qg.jpeg</url>
            <title>Stories by Aravind C on Medium</title>
            <link>https://medium.com/@aravindc42?source=rss-9dfee954f9db------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 02:46:33 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@aravindc42/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 Chatbot: Why TypeSafe AI’s Jev is Changing How Software Interacts with AI]]></title>
            <link>https://aravindc42.medium.com/beyond-the-chatbot-why-typesafe-ais-jev-is-changing-how-software-interacts-with-ai-bdeec5fafd01?source=rss-9dfee954f9db------2</link>
            <guid isPermaLink="false">https://medium.com/p/bdeec5fafd01</guid>
            <category><![CDATA[tech]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[agentic-rag]]></category>
            <category><![CDATA[jevs]]></category>
            <dc:creator><![CDATA[Aravind C]]></dc:creator>
            <pubDate>Sun, 04 Oct 2026 16:24:03 GMT</pubDate>
            <atom:updated>2026-10-04T16:24:03.090Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WxDRPqzUiq5dnY5G0pg-mw.png" /></figure><h3>How “System One” decision models eliminate JSON parsing headaches, sub-second latency bottlenecks, and runaway LLM costs.</h3><h3>Executive Summary</h3><ul><li><strong>The Problem:</strong> Traditional Large Language Models (LLMs) are over-engineered for simple control decisions. Asking GPT-4 or Claude to decide if a ticket is urgent takes 2 seconds, costs real money, and can break when JSON parsing fails.</li><li><strong>The Solution:</strong> <strong>Jev</strong> (launched September 2026 by TypeSafe AI) is not a chatbot. It is a <strong>System One decision model</strong> that evaluates input state and returns typed choices, scores, and probabilities in parallel.</li><li><strong>Key Specs:</strong> <strong>$0.042 per million input tokens</strong> (output is free), <strong>70–300 ms latency</strong>, and <strong>zero schema hallucinations</strong> (it physically cannot return a value outside your defined schema).</li><li><strong>The Takeaway:</strong> Use Jev alongside your LLMs for routing, guardrails, and verification — not for generating text or prose.</li></ul><h3>1. The Friction: Why Your Software Doesn’t Want Another Chatbot</h3><p>If you are a developer building AI-powered applications, you’ve likely run into this frustrating paradox:</p><blockquote><strong><em>You don’t need 70 billion parameters of literary creative writing just to decide if an incoming user prompt is malicious, or which team should own a support ticket.</em></strong></blockquote><p>When you ask a standard generative LLM for a binary choice or routing decision using structured output (JSON mode), three things happen:</p><ol><li><strong>Latency lag:</strong> The model auto-regressively generates tokens one by one, adding <strong>1.5 to 3 seconds</strong> of delay to your pipeline.</li><li><strong>Parsing fragility:</strong> Even with strict schemas, LLMs occasionally return markdown fences, trailing commas, or markdown commentary, causing runtime JSONDecodeError exceptions in production.</li><li><strong>Cost inefficiency:</strong> You pay for output generation on every request — even when the answer is just &quot;YES&quot; or &quot;NO&quot;.</li></ol><h3>2. The Three Primitives: How Jev Communicates</h3><p>Instead of prompts, Jev uses three deterministic questions called <strong>primitives</strong>:</p><ol><li><strong>Choice:</strong> Selects exactly <em>one</em> option from a list you define (e.g., categories: [&quot;billing&quot;, &quot;technical&quot;, &quot;spam&quot;]).</li><li><strong>Score:</strong> Rates input on a descriptive rubric scale (e.g., urgency level 1 to 5).</li><li><strong>Noul (Yes/No Probability):</strong> Evaluates a proposition and returns a calibrated probability between 0.0 and 1.0 (e.g., <em>&quot;Does this text contain prompt injection?&quot;</em>).</li></ol><h3>3. Final Takeaway</h3><p>Jev represents a shift in modern AI architecture. Instead of treating AI as a human conversational partner, Jev treats AI as a <strong>high-speed, typed microservice function</strong>.</p><p>If you are building AI agents, RAG pipelines, or customer support automation, don’t waste seconds and money making a massive LLM answer yes/no questions. Let a System One model make the decision in 100ms, and keep your large LLM reserved for what it does best: crafting human-level content</p><p>⭐ Thanks for reading this article ⭐</p><p>If I got something wrong? Let me know in the comments. I would love to improve.</p><p>Clap 👏 if this article helps you.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bdeec5fafd01" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Beyond the Chatbot: Why TypeSafe AI’s Jev is Changing How Software Interacts with AI]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://aravindc42.medium.com/beyond-the-chatbot-why-typesafe-ais-jev-is-changing-how-software-interacts-with-ai-d45e876b7d2b?source=rss-9dfee954f9db------2"><img src="https://cdn-images-1.medium.com/max/1365/1*WxDRPqzUiq5dnY5G0pg-mw.png" width="1365"></a></p><p class="medium-feed-snippet">How &#x201C;System One&#x201D; decision models eliminate JSON parsing headaches, sub-second latency bottlenecks, and runaway LLM costs.</p><p class="medium-feed-link"><a href="https://aravindc42.medium.com/beyond-the-chatbot-why-typesafe-ais-jev-is-changing-how-software-interacts-with-ai-d45e876b7d2b?source=rss-9dfee954f9db------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://aravindc42.medium.com/beyond-the-chatbot-why-typesafe-ais-jev-is-changing-how-software-interacts-with-ai-d45e876b7d2b?source=rss-9dfee954f9db------2</link>
            <guid isPermaLink="false">https://medium.com/p/d45e876b7d2b</guid>
            <category><![CDATA[agentic-rag]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[jevs]]></category>
            <category><![CDATA[llm]]></category>
            <dc:creator><![CDATA[Aravind C]]></dc:creator>
            <pubDate>Sun, 04 Oct 2026 16:22:20 GMT</pubDate>
            <atom:updated>2026-10-04T16:22:20.611Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The Prompt Bloat Paradox]]></title>
            <link>https://aravindc42.medium.com/the-prompt-bloat-paradox-fea930e0d47b?source=rss-9dfee954f9db------2</link>
            <guid isPermaLink="false">https://medium.com/p/fea930e0d47b</guid>
            <category><![CDATA[writing-prompts]]></category>
            <category><![CDATA[architecture]]></category>
            <category><![CDATA[best-practices]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ai-agent]]></category>
            <dc:creator><![CDATA[Aravind C]]></dc:creator>
            <pubDate>Fri, 17 Jul 2026 04:51:50 GMT</pubDate>
            <atom:updated>2026-07-17T04:51:50.278Z</atom:updated>
            <content:encoded><![CDATA[<p>It’s a scenario almost every AI engineer, developer, or builder knows all too well. You develop a prompt-based agent. You give it a few strong, foundational instructions, run a few dry runs with test cases, and it performs beautifully. The logic is sharp, the execution is spot-on, and the output is exactly what you envisioned. Success.</p><p>Naturally, you decide to take it to the next level. You want to turn this successful prototype into a comprehensive, production-ready, end-to-end use case. To do this, you start appending guardrails, processing logic, edge-case handlers, and output constraints.</p><p>Then, you hit run using the <em>exact same input</em> that worked perfectly ten minutes ago.</p><p>Suddenly, the magic is gone. The output is entirely different, core rules are inexplicably ignored, and the formatting is broken. It is as if your agent underwent a sudden, unexplained lobotomy.</p><p>Welcome to the <strong>Prompt Bloat Paradox</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*xabq8i6RmUzYZqoT" /><figcaption>Photo by <a href="https://unsplash.com/@lubean96?utm_source=medium&amp;utm_medium=referral">Lu Quang Do</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>Why “Just One More Rule” Breaks the System</h3><p>As builders, we tend to treat Large Language Models (LLMs) like traditional software scripts — assuming that adding one line of instruction is just like adding another if/else conditional statement to a Python script. But LLMs are probabilistic engines, not deterministic processors. When we over-complicate a single system prompt, we run headfirst into three architectural bottlenecks:</p><h3>1. Self-Attention Dilution</h3><p>The self-attention mechanism in modern transformer architectures allocates mathematical “weight” across all incoming tokens. If your system prompt is 100 words long, a crucial core instruction gets a high share of attention. If you bloat that prompt to 1,000 words to cover an entire end-to-end pipeline, the relative attention weight allocated to your core logic drops exponentially. The model isn’t actively defying you; your voice is simply lost in its own background noise.</p><h3>2. The “Lost in the Middle” Phenomenon</h3><p>Empirical research has repeatedly shown that language models are highly sensitive to information placed at the very beginning (primacy effect) and the very end (recency effect) of a prompt block. When you insert a web of complex business rules into the middle of a massive system prompt, the model’s retrieval capability for those specific constraints decays rapidly.</p><h3>3. Cognitive Contradiction Creep</h3><p>When prompts grow organically, we rarely rewrite them from scratch. Instead, we patch them. <em>“If X happens, do Y. But wait, if Z is true, ignore Y and do W.”</em> These overlapping, nested logic blocks often contain subtle contradictions that confuse the model’s internal statistical probabilities, causing it to default to an unexpected, chaotic fallback behavior.</p><blockquote><em>“An agent that is responsible for everything is ultimately responsible for nothing. Trying to make a single prompt handle an entire enterprise pipeline is the fastest path to regression.”</em></blockquote><h3>The Cure: Shifting from Monoliths to Micro-Agents</h3><p>If you are facing this issue, the solution is not to write a more clever, long-winded prompt. The solution is to change your architecture. We must apply a classic software engineering principle to AI: <strong>the Separation of Concerns</strong>.</p><p>Instead of forcing a single, overloaded agent to execute your entire end-to-end use case, you must decouple the workflow into a <strong>modular multi-agent pipeline</strong>. Each agent should have one highly-specific job, a minimized context window, and its own target objective.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vmy1I323pUZw6gvlnjhYdw.png" /></figure><h3>Step 1: Structural Clean-up</h3><p>If your framework cannot yet support multiple independent agents and you must keep it single-agent for now, you have to refactor your monolith prompt using strict XML-style tags or clear Markdown hierarchies. This signals to the model’s attention headers exactly how to segment and prioritize information.</p><pre># ROLE &amp; CORE GOAL<br>You are a specialized Data Solution Agent...<br><br># WORKFLOW STEPS<br>1. Analyze the input against target rules.<br>2. Formulate the solution payload.<br><br>&lt;guardrails&gt;<br>- Never alter the original keys.<br>- Do not add conversational filler.<br>&lt;/guardrails&gt;</pre><h3>Step 2: Implement the “Reviewer” Loop</h3><p>One of the most effective ways to restore dry-run quality in a complex chain is to introduce an explicit evaluation step. If a single agent struggles to handle the logic and format perfectly at the same time, pass its draft to a lighter, incredibly strict <strong>Evaluator Agent</strong> whose only instruction is: <em>“Compare this draft output with the original core instructions. Correct any deviations. Do not generate new information.”</em></p><h3>Engineering for Predictability</h3><p>As the generative AI space matures, we must move past the “magic prompt” phase of development. If your agent is failing because you tried to expand its scope into a full end-to-end workflow, take it as a sign of progress. It means you have outgrown the capabilities of a single system prompt, and it is time to build a robust system architecture.</p><p>Keep your agents small, their instructions focused, and their pipelines modular. Your production outcomes — and your peace of mind — will thank you.</p><p>⭐ Thanks for reading this article ⭐</p><p>If I got something wrong? Let me know in the comments. I would love to improve.</p><p>Clap 👏 if this article helps you.</p><h3>Read my other Blogs:</h3><ul><li><a href="https://aravindc42.medium.com/introduction-to-vector-search-in-google-cloud-gcp-afae160551e3">Introduction to Vector Search in Google Cloud (GCP)</a></li><li><a href="https://aravindc42.medium.com/itera-ai-%EF%B8%8F-how-i-built-a-one-stop-ai-powered-travel-planner-using-gemini-and-cloud-run-cloud-660792c3f4ed">Itera AI✈️: How I Built a One-Stop, AI-Powered Travel Planner Using Gemini and Cloud Run | Cloud…</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fea930e0d47b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Introduction to Vector Search in Google Cloud (GCP)]]></title>
            <link>https://aravindc42.medium.com/introduction-to-vector-search-in-google-cloud-gcp-afae160551e3?source=rss-9dfee954f9db------2</link>
            <guid isPermaLink="false">https://medium.com/p/afae160551e3</guid>
            <category><![CDATA[vector-database]]></category>
            <category><![CDATA[vector-search]]></category>
            <category><![CDATA[google-agentspace]]></category>
            <category><![CDATA[gcp]]></category>
            <category><![CDATA[embedding]]></category>
            <dc:creator><![CDATA[Aravind C]]></dc:creator>
            <pubDate>Mon, 25 May 2026 13:51:39 GMT</pubDate>
            <atom:updated>2026-05-25T14:41:26.820Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zo_0zugaHk3JGLZOe1Ibiw.png" /></figure><p>In the era of Generative AI and Large Language Models (LLMs), traditional keyword-based search is no longer enough. If you search for “chilly weather gear,” a traditional database might look for the exact words “chilly” or “weather”. But what you <em>actually</em> want are jackets, beanies, and sweaters.</p><p>This is where <strong>Vector Search</strong> comes into play. It enables computers to understand the deep, contextual meaning of data rather than just matching literal text.</p><p>In this guide, we will break down how Vector Search works under the hood and walk through a production-ready implementation using the Google Cloud AI Platform Python SDK.</p><h3>Core Concepts: What Happens Behind the Scenes?</h3><h4>1. Embeddings: Giving Data a “Mathematical Meaning”</h4><p>Computers don’t understand words, but they excel at math. An <strong>embedding</strong> is a process that converts an object (like a product name, text paragraph, image, or audio file) into a long list of numbers, known as a high-dimensional vector.</p><p>If two items are conceptually similar (e.g., “apple” and “green apple”), their vectors will be close to each other in this mathematical space.</p><h4>2. Approximate Nearest Neighbor (ANN) &amp; ScaNN</h4><p>If you have billions of items, comparing your search query vector to every single item in your database would take way too long.</p><p>To solve this, Google Cloud Vector Search uses the <strong>ScaNN </strong>algorithm. Instead of searching everything, it pre-clusters the vectors into a smart tree structure. When a query comes in, it instantly narrows the search down to the most relevant clusters, fetching results in mere <strong>milliseconds</strong>.</p><p>For additional details, please refer to this <a href="https://research.google/blog/announcing-scann-efficient-vector-similarity-search/">doc</a></p><h4>Index &amp; Index endpoint</h4><p>Building a semantic search engine on GCP involves three main structural components:</p><ul><li><strong>The Index:</strong> This is your core vector database. It holds all the embeddings and structures them using the ScaNN algorithm for quick retrieval.</li><li><strong>The Index Endpoint:</strong> Think of this as the server instance or API gateway. The index itself cannot be queried directly; it must be deployed to an endpoint, which listens for incoming search requests.</li></ul><h4>Prerequisites &amp; Setup</h4><p>Before we dive into the code, make sure you have a Google Cloud project set up with billing enabled.</p><p>This guide is heavily inspired by the official Google Cloud documentation. For the full architectural breakdown and initial environment setup, you can always refer back to the <a href="https://docs.cloud.google.com/gemini-enterprise-agent-platform/build/vector-search/quickstart">Google Cloud Gemini Enterprise Agent Platform Vector Search Quickstart</a>.</p><h4>Step-by-Step Code Walkthrough</h4><p>Let’s look at how to build and deploy a Vector Search system</p><p>You can follow along with this guide using either of these two environments, depending on your workflow and preferences:</p><ul><li><strong>Google Colab (Recommended for Quick Starts) </strong>or</li><li><strong>Gemini Enterprise Agent Platform Workbench (Recommended for Production)</strong></li></ul><p><em>Note: The full code for this walk-through is officially maintained by Google. You can view the complete interactive notebook on the </em><a href="https://github.com/GoogleCloudPlatform/generative-ai/blob/main/embeddings/vector-search-quickstart.ipynb"><em>Google Cloud Platform Generative AI GitHub Repository</em></a></p><h4>1. Prepare the data</h4><p>To build an index using the Gemini Enterprise Agent Platform, start by placing your embedding file into a Cloud Storage bucket.</p><h4>2. Create a index</h4><p>To kick off the configuration, you’ll need to define a few key parameters:</p><ul><li><strong>contents_delta_uri</strong>: The Cloud Storage directory path where your embedding JSON files are stored.</li><li><strong>dimensions</strong>: The size of each embedding vector. Since we are leveraging Google&#39;s Text Embeddings API for this project, set this to 768.</li><li><strong>approximate_neighbors_count</strong>: The number of similar items you want the model to retrieve during a typical search query.</li></ul><pre># init the aiplatform package<br>from google.cloud import aiplatform<br>aiplatform.init(project=PROJECT_ID, location=LOCATION)<br><br># create Index<br>my_index = aiplatform.MatchingEngineIndex.create_tree_ah_index(<br>    display_name = f&quot;vs-quickstart-index-{UID}&quot;,<br>    contents_delta_uri = BUCKET_URI,<br>    dimensions = 768,<br>    approximate_neighbors_count = 100,<br>)</pre><h4>3. Creating an Index Endpoint and Deploying</h4><p>To actually query your index, you need to host it on a server instance. In the Google Cloud ecosystem, this server instance is called an <strong>Index Endpoint</strong>. Think of it as the gateway that accepts your search queries and routes them to your vector data.</p><p>First, let’s spin up the endpoint:</p><pre># Create the Index Endpoint<br>my_index_endpoint = aiplatform.MatchingEngineIndexEndpoint.create(<br>    display_name=f&quot;vs-quickstart-index-endpoint-{UID}&quot;,<br>    public_endpoint_enabled=True<br>)</pre><p>Once your endpoint is live, you can deploy your index to it. You will need to specify a unique ID for this specific deployment:</p><pre># Define a unique deployment ID<br>DEPLOYED_INDEX_ID = f&quot;vs_quickstart_deployed_{UID}&quot;<br><br># Deploy the Index to your new Endpoint<br>my_index_endpoint.deploy_index(<br>    index=my_index, <br>    deployed_index_id=DEPLOYED_INDEX_ID<br>)</pre><blockquote><em>⏳ </em><strong><em>Take a Coffee Break:</em></strong><em> If this is your first time deploying this index, the backend setup can take around </em><strong><em>30 minutes</em></strong><em> to automatically build and initialize.</em></blockquote><p>If you want to track the deployment progress visually, head over to the Google Cloud Console. Navigate to the <strong>Agent Platform</strong> section, click on <strong>Vector search</strong>, and select <strong>Indexes</strong> to see the real-time status.</p><h4>4. Running a Query with Vector Search</h4><p>Now for the exciting part: testing our vector search. We will take a specific product name, grab its embedding vector, and ask our index endpoint to find the most similar products in our dataset.</p><h4>5. Load the Embeddings into Memory</h4><p>First, we need to load our local JSON file to map product IDs to their corresponding names and embedding vectors.</p><pre>import json<br># Initialize dictionaries to hold our product mapping<br>product_names = {}<br>product_embs = {}<br># Load and parse the embeddings file<br>with open(&#39;product-embs.json&#39;) as f:<br>    for line in f.readlines():<br>        product = json.loads(line)<br>        product_id = product[&#39;id&#39;]<br>        <br>        product_names[product_id] = product[&#39;name&#39;]<br>        product_embs[product_id] = product[&#39;embedding&#39;]</pre><p>With our data loaded, we can fetch the embedding vector for a target product. For this example, we’ll grab ID 6523 (which maps to <em>&quot;cloudveil women&#39;s excursion short&quot;</em>).</p><pre># Extract the query embedding for our target item<br># Feel free to experiment with other sample IDs like 12711, 18090, 19536, or 11863<br>query_emb = product_embs[&#39;6523&#39;]</pre><h4>6. Execute the Vector Search</h4><p>Next, we pass this vector to our endpoint using the find_neighbors() method. This will scan our index and return the closest matches based on semantic similarity.</p><pre># Run the query against our deployed index<br>response = my_index_endpoint.find_neighbors(<br>    deployed_index_id=DEPLOYED_INDEX_ID,<br>    queries=[query_emb],<br>    num_neighbors=10<br>)<br># Print out the top 10 most similar results<br>for idx, neighbor in enumerate(response[0]):<br>    print(f&quot;{neighbor.distance:.2f} {product_names[neighbor.id]}&quot;)</pre><h4>⚡ Why is this so fast?</h4><p>You’ll notice that the find_neighbors() method returns results in <strong>mere milliseconds</strong>.</p><p>Even if your production environment grows to hold billions of items, the search remains blindingly fast thanks to Google’s <strong>ScaNN </strong>algorithm.</p><h3>Conclusion</h3><p>Building an index with the <strong>Gemini Enterprise Agent Platform Vector Search</strong> unlocks the ability to search through billions of data points with millisecond latency, all backed by Google’s powerful ScaNN algorithm and seamless autoscaling.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=afae160551e3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Airflow 3 XComs: Passing Data Between Tasks (The Quick Guide)]]></title>
            <link>https://aravindc42.medium.com/airflow-3-xcoms-passing-data-between-tasks-the-quick-guide-aa269e42482c?source=rss-9dfee954f9db------2</link>
            <guid isPermaLink="false">https://medium.com/p/aa269e42482c</guid>
            <category><![CDATA[airflow]]></category>
            <category><![CDATA[etl]]></category>
            <category><![CDATA[xcom]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[pipeline]]></category>
            <dc:creator><![CDATA[Aravind C]]></dc:creator>
            <pubDate>Sat, 22 Nov 2025 15:37:55 GMT</pubDate>
            <atom:updated>2025-11-22T15:37:55.181Z</atom:updated>
            <content:encoded><![CDATA[<p>If you use Apache Airflow, you need a way for one task’s output to become the next task’s input. This mechanism is called <strong>XCom</strong> (Cross-Communication). Think of it as a small, high-speed messenger service between your tasks.</p><p><strong>XCom Golden Rule:</strong> Use it for <strong>small data</strong> (paths, IDs, summary statistics, small lists), not for huge datasets.</p><p>Lets see an example</p><h3>The Sales Aggregation Example</h3><p>We’ll use two PythonOperator tasks: one to <strong>push</strong> a year of sales data, and one to <strong>pull</strong> and calculate the average.</p><pre>from __future__ import annotations<br><br>import pendulum<br><br>from airflow.models.dag import DAG<br>from airflow.operators.python import PythonOperator<br><br># --- Define Sales Data ---<br># Representing sales data for 12 months (or any 12 periods)<br>SALES_DATA_1_YEAR = [<br>    12000, 15000, 18000, 20000, 14000, 16000,<br>    22000, 25000, 17000, 19000, 21000, 23000<br>]<br><br># --- 1. Python Function to Push Data (Producer) ---<br>def push_sales_data(**kwargs):<br>    &quot;&quot;&quot;<br>    Pushes the SALES_DATA_1_YEAR array into XCom.<br>    The &#39;return&#39; value of a python_callable is automatically pushed to XCom.<br>    &quot;&quot;&quot;<br>    print(f&quot;--- Pushing {len(SALES_DATA_1_YEAR)} sales values to XCom... ---&quot;)<br>    return SALES_DATA_1_YEAR<br><br># --- 2. Python Function to Pull Data and Aggregate (Consumer) ---<br>def aggregate_and_average_sales(ti):<br>    &quot;&quot;&quot;<br>    Pulls the sales data from the previous task, aggregates it,<br>    and calculates the average sales.<br>    &quot;&quot;&quot;<br>    # 1. Pull Data from XCom<br>    # ti.xcom_pull is used to retrieve data pushed by a previous task.<br>    # task_ids=&#39;push_data&#39; specifies which task to pull from.<br>    sales_array = ti.xcom_pull(task_ids=&#39;push_data&#39;)<br><br>    if not sales_array:<br>        print(&quot;Error: Could not retrieve sales data from XCom.&quot;)<br>        return<br><br>    # 2. Aggregate and Calculate Average<br>    total_sales = sum(sales_array)<br>    num_months = len(sales_array)<br>    average_sales = total_sales / num_months<br><br>    # 3. Print Results<br>    print(f&quot;\n--- Aggregation Results ---&quot;)<br>    print(f&quot;Total Sales (Sum): ${total_sales:,.2f}&quot;)<br>    print(f&quot;Number of Periods: {num_months}&quot;)<br>    print(f&quot;Average Sales: **${average_sales:,.2f}**&quot;)<br>    print(&quot;---------------------------\n&quot;)<br><br># --- DAG Definition ---<br>with DAG(<br>    dag_id=&quot;xcom_sales_aggregation_example&quot;,<br>    schedule=None,<br>    start_date=pendulum.datetime(2025, 1, 1, tz=&quot;UTC&quot;),<br>    catchup=False,<br>    tags=[&quot;xcom&quot;, &quot;sales&quot;, &quot;analytics&quot;],<br>) as dag:<br>    # Task 1: Push the array of sales data<br>    push_task = PythonOperator(<br>        task_id=&quot;push_data&quot;,<br>        python_callable=push_sales_data,<br>    )<br><br>    # Task 2: Pull the data and perform aggregation/average calculation<br>    aggregate_task = PythonOperator(<br>        task_id=&quot;calculate_average_sales&quot;,<br>        # We need to pass the context dictionary (including &#39;ti&#39;) to the function<br>        python_callable=aggregate_and_average_sales,<br>    )<br><br>    # --- Define Task Dependency ---<br>    push_task &gt;&gt; aggregate_task</pre><h3>Visualizing the Flow</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-rjDArQcTI7mnv-Bb3YB7w.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*poQpcCA1z7YqSe1l4__-sQ.jpeg" /></figure><h3>How XComs Work in This DAG</h3><h3>1. Pushing Data (The Producer)</h3><ul><li>In push_sales_data, we simply <strong>return</strong> the list SALES_DATA_1_YEAR.</li><li>The PythonOperator automatically captures this return value and stores it in the Airflow database as an <strong>XCom</strong> with the key return_value.</li></ul><h3>2. Pulling Data (The Consumer)</h3><ul><li>The aggregate_and_average_sales function receives the <strong>Task Instance (</strong><strong>ti)</strong> object.</li><li>We use the ti.xcom_pull(task_ids=&#39;push_data&#39;) method:</li><li><strong>ti.xcom_pull</strong>: The method used to retrieve XCom data.</li><li><strong>task_ids=&#39;push_data&#39;</strong>: Specifies which task we want to retrieve the data <em>from</em>.</li></ul><p>This two-step process allows the sales array to safely move from the first task’s execution environment to the second, enabling the final calculation.</p><h3>Key Takeaways</h3><ul><li>Xcom is a mechanism for tasks to exchange <strong>small amounts of data</strong>.<strong>Data Rule</strong></li><li>Use for <strong>small data</strong> (paths, IDs). <strong>Do not</strong> use for large files/datasets.<strong>Pushing Data</strong></li><li>The return value of a PythonOperator function is <strong>automatically</strong> pushed to XCom (return_value key).<strong>Pulling Data</strong>Use the <strong>Task Instance (</strong><strong>ti)</strong> object and the method <strong>ti.xcom_pull(task_ids=&#39;...&#39;)</strong> to retrieve data from a previous task.</li><li>The producer task must run and <strong>succeed</strong> before the consumer task attempts to pull the data.</li></ul><h3>For More Information</h3><p>To explore more advanced XCom topics, including using custom keys and overriding default behavior, check out the official documentation:</p><p><strong>Apache Airflow Documentation:</strong> <a href="https://airflow.apache.org/docs/apache-airflow/stable/core-concepts/xcoms.html">https://airflow.apache.org/docs/apache-airflow/stable/core-concepts/xcoms.html</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=aa269e42482c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Itera AI✈️: How I Built a One-Stop, AI-Powered Travel Planner Using Gemini and Cloud Run | Cloud…]]></title>
            <link>https://aravindc42.medium.com/itera-ai-%EF%B8%8F-how-i-built-a-one-stop-ai-powered-travel-planner-using-gemini-and-cloud-run-cloud-660792c3f4ed?source=rss-9dfee954f9db------2</link>
            <guid isPermaLink="false">https://medium.com/p/660792c3f4ed</guid>
            <category><![CDATA[google-cloud-run]]></category>
            <category><![CDATA[gemini]]></category>
            <category><![CDATA[hackathons]]></category>
            <category><![CDATA[google-ai-studio]]></category>
            <category><![CDATA[cloud-run-hackathon]]></category>
            <dc:creator><![CDATA[Aravind C]]></dc:creator>
            <pubDate>Sun, 09 Nov 2025 19:29:57 GMT</pubDate>
            <atom:updated>2025-11-09T19:31:28.039Z</atom:updated>
            <content:encoded><![CDATA[<h3>Itera AI✈️: How I Built a One-Stop, AI-Powered Travel Planner Using Gemini and Cloud Run | Cloud run hackathon 2025</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*QkiucdIIE41kLSDw.jpeg" /></figure><h3>Inspiration: From Fragmented Frustration to a Unified Vision</h3><p>As a dedicated <strong>wanderlust</strong> with a passion for exploring new corners of the globe, I’ve always cherished the thrill of travel. However, I consistently faced a significant hurdle before every trip: the overwhelming and fragmented process of planning. Crafting the perfect itinerary meant juggling countless browser tabs and mobile apps. I’d have one app for weather forecasts, another for hotel bookings, several blogs and travel sites for activity ideas, and mapping services to stitch it all together. The time spent on planning — researching climate, budgeting, finding restaurants, and creating a pre-trip shopping list — often felt like it rivaled the duration of the trip itself.</p><p>This personal frustration was the seed for <strong>Itera AI</strong>. I envisioned a single, intelligent platform that could consolidate all these scattered pieces of information into one cohesive, personalized plan. I wanted to build a tool, powered by the generative AI capabilities of <strong>Gemini</strong>, where a traveler could get everything they need in one place, transforming hours of stressful research into a moment of joyful anticipation.</p><h3>What It Does: Your AI Travel Agent</h3><p>Itera AI is a sophisticated, end-to-end travel planning application designed to eliminate the complexity and time-consuming nature of creating a personalized trip itinerary. It acts as a smart travel assistant, leveraging the power of the <strong>Google Gemini API</strong> to generate a comprehensive and highly detailed travel plan from a single set of user preferences.</p><h3>Core Functionality:</h3><ul><li><strong>Intuitive Preference Gathering:</strong> The user starts by filling out a clean form detailing their destination, travel dates, budget, preferred travel pace, and specific interests.</li><li><strong>AI-Powered Itinerary Generation:</strong> The application sends these preferences in a structured prompt to the Gemini API, requesting a complete travel plan in a specific JSON format.</li><li><strong>Day-by-Day Itinerary:</strong> Each day has a theme and a list of activities with suggested times, detailed descriptions, transport options, specific locations, and estimated costs tailored to the user’s budget and local currency.</li><li><strong>Enriching Visuals:</strong> The app makes subsequent calls to Gemini’s image generation model (gemini-2.5-flash-image) to create a unique, relevant photo for each activity.</li><li><strong>Essential Travel Intelligence:</strong> Beyond the schedule, the plan includes a weather forecast, advice on the best time to visit, safety tips, cultural etiquette, and even a personalized packing list.</li></ul><h3>Key Interactive Features:</h3><ul><li><strong>Activity Swapping:</strong> If a user doesn’t like a suggested activity, a <strong>“Swap” button</strong> triggers a targeted API call to Gemini, asking for a new activity that fits the context of the day and their original preferences. This creates a truly dynamic plan.</li><li><strong>Utility Tools:</strong> Users can download their complete, visually-rich itinerary as a <strong>PDF for offline use</strong> or use the “Share” feature to generate a clean, text-based summary.</li></ul><h3>Challenges faced</h3><p>The most significant challenge was engineering a truly resilient and error-free application. When working with a generative AI model, you must anticipate and gracefully handle a wide range of potential issues — from network failures and API rate limits to unexpected or malformed JSON responses from the model. My focus was on ensuring that no backend issue would compromise the customer experience. I implemented comprehensive error handling to catch API exceptions and provide the user with clear, helpful feedback. This process taught me the importance of building defensive, user-centric logic that maintains a high-quality experience even when things go wrong behind the scenes. Itera AI is a testament to the idea that a powerful application should not only be feature-rich but also <strong>robust and reliable</strong>.</p><h3>How We Built It: A Serverless, Gemini-Powered Architecture</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*sU5qHX6BYycdduZg.jpeg" /></figure><p><strong>Itera AI was created for the Cloud Run Hackathon and developed using Google AI Studio.</strong></p><p>Itera AI is architected as a serverless Single-Page Application (SPA) deployed on <strong>Google Cloud Run</strong> for seamless scalability and management. The user’s browser runs the entire application (built with <strong>React</strong>), while all the complex logic, content, and image generation are handled directly by the <strong>Google Gemini API</strong>, which acts as a powerful, on-demand “backend.”</p><h3>1. The “AI Backend”: The Gemini Service Layer</h3><p>The core intelligence resides in a dedicated service layer that encapsulates all communication with the Gemini API.</p><ul><li><strong>Structured Output with </strong><strong>responseSchema:</strong> This was the most critical technique for reliability. Instead of asking the AI for plain text, the generateItinerary function provides a detailed schema. This <strong>forces the </strong><strong>gemini-2.5-flash model to return its response in a predictable, perfectly structured JSON format</strong>, which is the key to consistency.</li><li><strong>Multi-Model Strategy:</strong> The service intelligently uses different Gemini models for different tasks:</li><li><strong>gemini-2.5-flash:</strong> Used for all text and logic-based generation (the main itinerary, destination validation, activity swapping) because of its excellent speed and reasoning capabilities.</li><li><strong>gemini-2.5-flash-image:</strong> Used to create rich, relevant visuals for each activity in the plan, making the itinerary more engaging.</li></ul><h3>2. Deployment: Serving the App with Google Cloud Run</h3><p>Deploying a React SPA on Cloud Run is an ideal choice for a serverless, scalable, and cost-effective solution.</p><ul><li><strong>Containerization with Docker:</strong> A Dockerfile uses a multi-stage build: a Node image to compile the React code into static files, and a second, lightweight <strong>Nginx</strong> image to efficiently serve those files.</li><li><strong>Deployment:</strong> The final container image is stored in <strong>Google Artifact Registry</strong> and then deployed to a <strong>Cloud Run</strong> service, creating a highly available and auto-scaling production environment.</li></ul><h3>🌟 Additional Features</h3><p>Our development of Itera AI yielded several key achievements that underscore the power of combining Google’s AI and Cloud infrastructure:</p><ul><li><strong>Dynamic and Interactive Plans:</strong> We went beyond simple static text generation by implementing the AI-powered <strong>“Swap Activity”</strong> feature. This is a testament to the Gemini model’s advanced reasoning capabilities, allowing it to instantly generate a relevant new suggestion that fits the context of the day and the user’s original travel preferences. This is what truly transforms a plan from a document into a dynamic, personalized tool.</li><li><strong>Multi-Modal and Efficient Architecture:</strong> We intelligently utilized a multi-model strategy for maximum efficiency and engagement. We relied on the <strong>gemini-2.5-flash</strong> model for all fast, logic-based tasks like the main itinerary and activity swapping, and leveraged the <strong>gemini-2.5-flash-image</strong> model to create rich, unique visuals for each activity. This blend of text and image generation makes the final itinerary highly engaging and demonstrates an efficient use of the Gemini API suite.</li><li><strong>Serverless and Scalable Deployment:</strong> By architecting Itera AI as a modern, containerized Single-Page Application (SPA) deployed entirely on <strong>Google Cloud Run</strong>, we created a solution that is inherently scalable, cost-effective, and fully serverless. This robust deployment pipeline — which includes <strong>Docker</strong> and <strong>Artifact Registry</strong> — ensures the application can handle travelers globally without manual server management.</li></ul><h3>What’s Next for Itera AI</h3><p><strong>Real-Time Collaboration</strong>: Allowing users to plan trips together with friends and family. Live Flight &amp; Hotel Booking: Integrating booking platforms to make Itera AI a one-stop shop.</p><p><strong>AI-Powered Travel Agent Chat:</strong> A conversational interface for making changes and asking questions. Budget Tracking &amp; Expense Management: Tools to help travelers stay on budget during their trip.</p><p><strong>Check out this demo video of Itera AI : </strong><a href="https://www.youtube.com/watch?v=10pSARDdihg">https://www.youtube.com/watch?v=10pSARDdihg</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=660792c3f4ed" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Itera AI✈️: How I Built a One-Stop, AI-Powered Travel Planner Using AI Studio and Cloud Run |…]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://aravindc42.medium.com/itera-ai-%EF%B8%8F-how-i-built-a-one-stop-ai-powered-travel-planner-using-ai-studio-and-cloud-run-6a681a3d7bb2?source=rss-9dfee954f9db------2"><img src="https://cdn-images-1.medium.com/max/2600/1*IKk72UsR0LBTeAZ8S8Z6pw.jpeg" width="2862"></a></p><p class="medium-feed-snippet">Inspiration: From Fragmented Frustration to a Unified Vision</p><p class="medium-feed-link"><a href="https://aravindc42.medium.com/itera-ai-%EF%B8%8F-how-i-built-a-one-stop-ai-powered-travel-planner-using-ai-studio-and-cloud-run-6a681a3d7bb2?source=rss-9dfee954f9db------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://aravindc42.medium.com/itera-ai-%EF%B8%8F-how-i-built-a-one-stop-ai-powered-travel-planner-using-ai-studio-and-cloud-run-6a681a3d7bb2?source=rss-9dfee954f9db------2</link>
            <guid isPermaLink="false">https://medium.com/p/6a681a3d7bb2</guid>
            <category><![CDATA[hackathons]]></category>
            <category><![CDATA[google-cloud-run]]></category>
            <category><![CDATA[cloud-run-hackathon]]></category>
            <category><![CDATA[google-ai-studio]]></category>
            <category><![CDATA[gemini]]></category>
            <dc:creator><![CDATA[Aravind C]]></dc:creator>
            <pubDate>Sun, 09 Nov 2025 19:25:04 GMT</pubDate>
            <atom:updated>2025-11-09T19:25:04.160Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Resources to prepare for your Microsoft Azure Fundamentals Exam]]></title>
            <link>https://aravindc42.medium.com/resources-to-prepare-for-your-microsoft-azure-fundamentals-exam-759fbac5daaf?source=rss-9dfee954f9db------2</link>
            <guid isPermaLink="false">https://medium.com/p/759fbac5daaf</guid>
            <category><![CDATA[microsoft]]></category>
            <category><![CDATA[cloud]]></category>
            <category><![CDATA[azure]]></category>
            <category><![CDATA[certification]]></category>
            <category><![CDATA[az900]]></category>
            <dc:creator><![CDATA[Aravind C]]></dc:creator>
            <pubDate>Sun, 15 Aug 2021 09:43:09 GMT</pubDate>
            <atom:updated>2021-08-15T09:43:09.993Z</atom:updated>
            <content:encoded><![CDATA[<p>Hey everyone, In this blog, I just want to share my experience about passing Microsoft Azure Fundamentals Exam (AZ-900) and the resources which I have used during my preparation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7Y9rvAsf3FtGnwoOz7H-kA.png" /></figure><p>Let’s dive right in.</p><p>These are the main resources you have to look at when you are about to get started:</p><h3>1. Microsoft Portal</h3><p>Docs are the best!! Microsoft has well structured the course and provided all the sufficient details about the exam. So at the first step go to the portal, download the course outline structure and make sure that you have a better understanding of what the exam covers.</p><p>So when it comes to AZ900 since this is the fundamentals exam it covers the basic knowledge about several cloud services provided by Microsoft Azure.</p><p><strong>Concepts covered:</strong></p><ol><li>Cloud concepts</li><li>Azure core services</li><li>Azure core solutions and management</li><li>Azure Security and network security features</li><li>Identity, governance, privacy, and compliance features</li><li>Azure cost management and Service Level Agreements</li></ol><p>For further additional details do visit: <a href="https://docs.microsoft.com/en-us/learn/certifications/exams/az-900">Exam AZ-900: Microsoft Azure Fundamentals — Learn</a></p><p>There are two types of training provided by Microsoft are Online-Free &amp; Instructor-led -Paid Course.</p><p>I completed Online-Free training which was useful and the course was structured based on the exam outline. And also along with the video courses it had some Multiple choice questions which you can test yourself.</p><h3>2. Adam Marczak — Azure for Everyone</h3><p>One of the popular and well-explained courses for AZ900 is Adam Marczak AZ900 Playlist which is available on youtube. The best part is after every concept you can find Prep Notes, Test, and more. So that at the end you can refresh all the concepts by using these resources.</p><p>Checkout: <a href="https://youtube.com/playlist?list=PLGjZwEtPN7j-Q59JYso3L4_yoCjj2syrM">Microsoft Azure Fundamentals (AZ-900) Full Course</a></p><h3>3.Udemy Course: AZ-900: Microsoft Azure Fundamentals Exam Prep — MAY 2021</h3><p>Another course that I can recommend is Microsoft Azure Fundamentals Exam Prep by Scott Duffy. The course duration is about 5hours which covers all the concepts as required in the exam. In addition to that, it has practice test questions by which you can test your knowledge.</p><p>Checkout: <a href="https://www.udemy.com/course/az900-azure/">AZ-900: Microsoft Azure Fundamentals Exam Prep — MAY 2021</a></p><p>This is more than enough to get prepared, so what’s next?</p><p><strong>Practice! Practice! Practice!</strong></p><p>Yes, the more you practice the better you become a pro!!</p><h3>4. WHIZLABS:</h3><p>Whizlabs is one of the top online training providers for certifications. So if you want to start from scratch or test yourself by attending mock tests then Whizlabs is the right place for you. I used Whizlabs for practicing Mock tests for AZ900 where it had 440+ Questions, Lab sessions, Online course which was useful and it gave me more confidence to crack the exam.</p><p>Visit for more info: <a href="https://www.whizlabs.com/microsoft-azure-certification-az-900/">Microsoft Azure Exam AZ-900 Certification</a></p><p>Yay! you made it. That’s all from my side. These are the resources that helped me to clear the Microsoft Azure Fundamentals exam. I hope this blog will give you the necessary resources to clear the exam.</p><p>Feel free to reach me if you have any doubts. Happy to help!</p><p><em>Originally published at </em><a href="https://techwiz.hashnode.dev/resources-to-prepare-for-your-microsoft-azure-fundamentals-exam-az-900"><em>https://techwiz.hashnode.dev</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=759fbac5daaf" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Among us game animation using Flutter]]></title>
            <link>https://aravindc42.medium.com/among-us-game-animation-using-flutter-91ef983707bc?source=rss-9dfee954f9db------2</link>
            <guid isPermaLink="false">https://medium.com/p/91ef983707bc</guid>
            <category><![CDATA[open-source]]></category>
            <category><![CDATA[animation]]></category>
            <category><![CDATA[flutter]]></category>
            <category><![CDATA[games]]></category>
            <category><![CDATA[among-us]]></category>
            <dc:creator><![CDATA[Aravind C]]></dc:creator>
            <pubDate>Sat, 15 May 2021 13:45:16 GMT</pubDate>
            <atom:updated>2021-05-15T13:45:16.983Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Zu4-y279riTeUC-_bxZvaA.png" /></figure><p>Hey, Flutter devs out there! Have you ever played a game and got attracted to the User interface, animations involved? As a developer have you wondered how those animations work?</p><p>A few months ago, I played a game called <strong>Among Us</strong> and loved the game context, characters, animations involved and thought of replicating the same using <strong>Flutter</strong>.</p><p>Guess what 💡 the result what as expected and I was able to produce the same using Flutter.</p><p>Let’s check out how to achieve this</p><h3>Table of Contents:</h3><ol><li>Introduction to Animation Controller</li><li>Source code implementation</li><li>Output</li><li>Source File</li><li>Conclusion</li></ol><h4>1. Introduction to Animation Controller</h4><p>Basically, the Animation controller helps you to control your animation object. It allows us to play an animation forward, backward.</p><p>By default, the animation controller ranges values from <strong>0.0 to 1.0</strong>. As per the doc, an AnimationController needs a TickerProvider, which is configured using the vsync argument on the constructor.</p><p>We’ll be using SingleTickerProviderStateMixin in our stateful widget.</p><p>For more info check out <a href="https://api.flutter.dev/flutter/animation/AnimationController-class.html">Flutter Docs</a></p><h4>2. Source code implementation</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*g6iZdnvlpB_RGRhVtHGnzA.gif" /></figure><p>This is our expected output, a floating character that rotates from left to right</p><p>Let’s start with the basic stateful widget named FloatingPage which contains a picture of an among us character in the center.</p><pre>class _FloatingPageState extends State&lt;FloatingPage&gt; {<br>  <a href="http://twitter.com/override">@override</a><br>  Widget build(BuildContext context) {<br>    return Scaffold(<br>      backgroundColor: Colors.black,<br>      body: Center(<br>        child: Container(<br>            height: 300, width: 200, child: Image.asset(&quot;images/green.png&quot;)),<br>      ),<br>    );<br>  }<br>}</pre><p><strong>Preview</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/435/0*3bOoRyX2IcD6ZmEc" /></figure><p>Now let’s add animation components, to achieve the expected output we have to add two things</p><p><strong>1. Translation</strong></p><p><strong>2. Rotation</strong></p><ul><li><strong>Creating an animation control object</strong></li></ul><pre>AnimationController animationControl;</pre><pre>animationControl = AnimationController(<br>      duration: Duration(seconds: 10),<br>      vsync: this,<br>    );</pre><ul><li><strong>For translation effect, let’s add a property called translation value</strong></li></ul><pre>Animation&lt;double&gt; traslationValue;</pre><pre>traslationValue = Tween&lt;double&gt;(<br> begin: 0,<br> end: 500,<br> ).animate(animationControl);</pre><p>Begin represents the starting point of the object and the end value represents the endpoint where the object gets stopped.</p><ul><li><strong>For rotation effect, let’s add a property called rotation value</strong></li></ul><pre>Animation&lt;double&gt; rotationValue;</pre><pre>rotationValue = Tween&lt;double&gt;(<br>      begin: 0,<br>      end: 2 * pi,<br>    ).animate(animationControl);</pre><p>Adding both translation value and rotation value to our character object.</p><pre>import &#39;dart:math&#39;;</pre><pre>import &#39;package:flutter/material.dart&#39;;</pre><pre>class FloatingPage extends StatefulWidget {<br>  <a href="http://twitter.com/override">@override</a><br>  _FloatingPageState createState() =&gt; _FloatingPageState();<br>}</pre><pre>class _FloatingPageState extends State&lt;FloatingPage&gt;<br>    with SingleTickerProviderStateMixin {<br>  AnimationController animationControl;</pre><pre>Animation&lt;double&gt; rotationValue;<br>  Animation&lt;double&gt; traslationValue;</pre><pre><a href="http://twitter.com/override">@override</a><br>  void initState() {<br>    // TODO: implement initState<br>    super.initState();<br>    animationControl = AnimationController(<br>      duration: Duration(seconds: 20),<br>      vsync: this,<br>    );</pre><pre>rotationValue = Tween&lt;double&gt;(<br>      begin: 0,<br>      end: 2 * pi,<br>    ).animate(animationControl);</pre><pre>traslationValue = Tween&lt;double&gt;(<br>      begin: 0,<br>      end: 500,<br>    ).animate(animationControl);</pre><pre>animationControl.forward();<br>  }</pre><pre><a href="http://twitter.com/override">@override</a><br>  Widget build(BuildContext context) {<br>    return Scaffold(<br>      backgroundColor: Colors.black,<br>      body: AnimatedBuilder(<br>        animation: animationControl,<br>        builder: (_, child) =&gt; Center(<br>          child: Stack(children: [<br>            Positioned(<br>                top: 100,<br>                left: traslationValue.value,<br>                child: Transform(<br>                    transform: Matrix4.rotationZ(rotationValue.value),<br>                    alignment: Alignment.center,<br>                    child: Image.asset(<br>                      &#39;assets/images/green.png&#39;,<br>                      height: 160,<br>                      width: 160,<br>                    ))),<br>          ]),<br>        ),<br>      ),<br>    );<br>  }</pre><pre><a href="http://twitter.com/override">@override</a><br>  void dispose() {<br>    // TODO: implement dispose<br>    animationControl.dispose();<br>    super.dispose();<br>  }<br>}</pre><h4>3. Output</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*F-LsP37qRMS0m0_-NfDaow.gif" /></figure><h4>4. Source code: <a href="https://github.com/aravindc5/amoung_us_character_animation">Github</a></h4><p>Do give a ⭐ if you find this interesting.</p><h4>5. Conclusion</h4><p>Hope this blog helps you to achieve the animation.</p><p>Did I get something wrong? Let me know in the comments. I would love to improve 💗</p><p>If you want any other animations in flutter let me know in the comments.</p><p>Like my work? You can buy me a coffee using the below link.</p><p><a href="https://www.buymeacoffee.com/Arvi">Aravind C is building Android apps, Blogs related to Programming &amp; Technology</a></p><h4>Read my other blogs</h4><ul><li><a href="https://medium.com/techvile/these-libraries-will-make-you-fall-in-love-with-flutter-techvile-6c727cc24a1d">These libraries will make you fall in love with Flutter💙 — Techvile</a></li><li><a href="https://medium.com/techvile/lets-make-list-animation-easy-using-flutter-90c4235e1d5c">Let’s make List Animation easy using Flutter!</a></li></ul><p><em>Originally published at </em><a href="http://techwiz.hashnode.dev/among-us-game-animation-using-flutter"><em>https://techwiz.hashnode.dev</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=91ef983707bc" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Become a DATA ANALYST with NO degree? The Google Data Analytics Professional Certificate.]]></title>
            <link>https://aravindc42.medium.com/google-data-analytics-certificate-overview-924757431c81?source=rss-9dfee954f9db------2</link>
            <guid isPermaLink="false">https://medium.com/p/924757431c81</guid>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[google]]></category>
            <category><![CDATA[datasciencecourse]]></category>
            <category><![CDATA[certification]]></category>
            <category><![CDATA[data]]></category>
            <dc:creator><![CDATA[Aravind C]]></dc:creator>
            <pubDate>Fri, 07 May 2021 06:33:33 GMT</pubDate>
            <atom:updated>2021-05-10T04:35:09.040Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*YDw7lG87vLTKnxBK" /><figcaption>Photo by <a href="https://unsplash.com/@lukechesser?utm_source=medium&amp;utm_medium=referral">Luke Chesser</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Hey, Aspiring learners out there! Ever wanted to start a career in the field of Data? Let’s check out the new professional certificate provided by google.</p><h3>Table of Contents:</h3><ul><li>Introduction</li><li>About the course</li><li>Tools &amp; Skills involved in the course</li><li>Benefits</li><li>Conclusion</li></ul><h3>INTRODUCTION</h3><p>Google recently announced a course for Data analytics in the month of March. This course aimed for the students or associates to get introduced to the field of Data analytics.</p><h4>What is Data Analytics?</h4><p>As the process of analyzing raw data to find trends and answer questions. Data Scientists and Analysts use data analytics techniques in their research, and businesses also use them to inform their decisions.</p><p>Data analysis can help companies better understand their customers, evaluate their ad campaigns, personalize content, create content strategies and develop products.</p><h3>ABOUT THE COURSE</h3><p>Google Data Analytics course is available on <a href="https://www.coursera.org/professional-certificates/google-data-analytics?utm_source=google&amp;utm_medium=institutions&amp;utm_campaign=gwgsite&amp;_ga=2.34437153.1294624843.1620214346-1385624920.1619259813">Coursera</a> and you can enroll at a cost of $39 per month for all eight classes. As such, the faster you complete the class, the better it is.</p><p>There are 8 Courses in this Professional Certificate:</p><h4>1. Foundations: Data, Data, Everywhere :</h4><p>This course describes the basic foundation of data and tells us the lifecycle and its functionality. And also it will cover day to day responsibility of a data analyst. At the end of this course, we can discover a wide variety of terms and concepts relevant to the role of a junior data analyst.</p><h4>2. Asking Questions to Make Data-Driven Decisions:</h4><p>This course describes effective questioning techniques that can help guide analysis. And also understanding of data-driven decision-making and how data analysts present findings.</p><h4>3. Preparing Data for Exploration</h4><p>This course helps to identify different types of bias in data to help ensure data credibility. And to find the relationship between and importance of data ethics and data privacy. It also covers best practices for organizing data and keeping it secure.</p><h4>4. Process Data from Dirty to Clean</h4><p>This course is one of the best because it covers different methods to process a huge number of clustered data to the precise data, where it involves various methods of data cleaning and methods, logic to apply them. It covers tools like both SQL and Excel sheets.</p><h4>5. Analyze Data to Answer Questions</h4><p>This course describes what is involved in the data analysis process with reference to goals and key tasks, and the importance of organizing data before analysis with references to sorts and filters.</p><h4>6. Share Data Through the Art of Visualization</h4><p>This course explains the key concepts involved in design thinking as they relate to data visualization. And the use of data visualizations to talk about data and the results of data analysis, also the key concepts involved in data visualization.</p><h4>7. Data Analysis with R Programming</h4><p>This course describes the RStudio programming environment including its components and benefits, and programming languages and appropriate use including examples. It also differentiates between the R Console and R programming environments.</p><h4>8. Capstone: Complete a Case Study</h4><p>A case study deals with the basic understanding of previous courses, where we have to apply those concepts in such a way that at the end of the study we should be able to make data-driven decisions. Since it’s an optional one but it is highly recommended to take participation to have some hands-on training and gain some confidence towards data analytics.</p><h3>TOOLS &amp; SKILLS</h3><p>These are the tools that are covered in the course:</p><h4>1. Tableau Software</h4><p>Tableau is the fastest-growing data visualization and data analytics tool that aims to help people see and understand data.</p><p>In other words, it simply converts raw data into a very easily understandable format. Data analysis is great, as it is a powerful visualization tool in the business intelligence industry.</p><h4>2. R Programming</h4><p>R is a programming language and software environment for statistical analysis, graphics representation, and reporting.</p><p>Uses:</p><ul><li>Data Analytics.</li><li>Statistical inferences.</li><li>Machine learning and deep learning.</li></ul><h4>3. Google Big Query</h4><p>BigQuery is a fully managed, serverless data warehouse that enables scalable analysis over petabytes of data. BigQuery was designed for analyzing data on the order of billions of rows, using a SQL-like syntax.</p><h3>BENEFITS</h3><p>Google has designed this course to prepare you for a job in your field of choice, you’ll get access to a resume building tool, mock interviews, and career networking support designed to help you with your job search. You’ll also be able to connect with over 130 US employers in the hiring consortium who are accepting candidates who have completed a Google Career Certificate.</p><h3>CONCLUSION</h3><p>I recently completed this course and had a very good experience so if you guys have some doubts do let me know in the comments or reach me at <a href="https://www.linkedin.com/in/aravindc1999/">LinkedIn</a>, <a href="https://twitter.com/arvii_14">Twitter</a>.</p><p>Check out my other blogs at <a href="https://medium.com/@aravindc42">Medium</a></p><p>Hope this blog helps you to get started in the field of Data Analytics. Do share your views. 😄</p><p><em>Originally published at </em><a href="https://techwiz.hashnode.dev/become-a-data-analyst-with-no-degree-the-google-data-analytics-professional-certificate"><em>https://techwiz.hashnode.dev</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=924757431c81" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>