<?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 Vijayavel Palaniappan on Medium]]></title>
        <description><![CDATA[Stories by Vijayavel Palaniappan on Medium]]></description>
        <link>https://medium.com/@vijayavel.p?source=rss-33350d5fc5f2------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*gKCZAoUnpCO_mFte.</url>
            <title>Stories by Vijayavel Palaniappan on Medium</title>
            <link>https://medium.com/@vijayavel.p?source=rss-33350d5fc5f2------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 22:18:20 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@vijayavel.p/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[From One Prompt to Five AI Agents: Building AI Newsletter Daily]]></title>
            <link>https://medium.com/@vijayavel.p/from-one-prompt-to-five-ai-agents-building-ai-newsletter-daily-6e20c3b8626d?source=rss-33350d5fc5f2------2</link>
            <guid isPermaLink="false">https://medium.com/p/6e20c3b8626d</guid>
            <category><![CDATA[multi-agent-systems]]></category>
            <category><![CDATA[prompt-engineering]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[python3]]></category>
            <category><![CDATA[ai-orchestration]]></category>
            <dc:creator><![CDATA[Vijayavel Palaniappan]]></dc:creator>
            <pubDate>Mon, 19 Oct 2026 06:02:20 GMT</pubDate>
            <atom:updated>2026-10-19T06:04:33.399Z</atom:updated>
            <content:encoded><![CDATA[<p>Few weeks ago, I started with a simple Python script that generated a newsletter and sent it to my email. Today, five AI agents from different providers research, write, review, and deliver a five-minute tech newsletter to my inbox at 5 AM every morning without any manual intervention.</p><p>What started as a small experiment has evolved into a multi-agent AI system. Here are the challenges I faced, the design decisions I made, and the lessons I learned along the way.</p><h3>1. Starting Small: One Agent and Several Setup Issues</h3><p>My first version had one AI model, a simple prompt, and an email tool. The model generated the newsletter, while a separate module handled email delivery, recipient validation, and approval.</p><p>Getting the first email delivered took longer than writing the code. I ran into environment configuration issues, Gmail App Password authentication, SSL certificates, missing API credits, and VS Code settings.</p><p>The biggest lesson was to test each component independently before debugging the entire application. Simple SMTP checks and dry runs helped me identify configuration problems quickly.</p><h3>2. The Problem with AI News Without Web Search</h3><p>Initially, I asked the agent to generate a newsletter covering the latest AI and software architecture news. The content looked convincing, but the agent had no access to current information.</p><p>It was relying on its existing knowledge, which meant stories could be outdated or inaccurate. I added web search and instructed the agent to use only stories found in search results, with links to their sources.</p><p><strong>Asking an AI model for the latest information does not mean it has access to it. The agent needs the right tools to do the job.</strong></p><h3>3. Automating the Newsletter</h3><p>I moved the application to GitHub Actions so it could run every morning without keeping my laptop switched on.</p><p>A few important details needed attention:</p><ul><li><strong>Scheduling:</strong> GitHub Actions uses UTC, so I configured schedules for both daylight saving and standard time to target 5 AM Pacific.</li><li><strong>Email safety:</strong> I disabled interactive approval in CI and retained recipient validation and a one-email-per-run limit.</li><li><strong>Failure notifications:</strong> If the newsletter cannot be delivered, the workflow fails visibly instead of silently leaving my inbox empty.</li></ul><p>Automation is not just about scheduling a script. It also means knowing when something goes wrong.</p><h3>4. Moving to a Multi-Agent Architecture</h3><p>To improve content quality and diversity, I redesigned the application as a small newsroom with four writer agents and one editor.</p><p>Instead of giving every writer a separate search implementation, I built one shared research pipeline that collects approximately 90 articles from the previous 24 hours. All four writers receive the same numbered list and reference stories by number. The application generates the source links, reducing the risk of fabricated URLs.</p><p>The editor evaluates relevance, accuracy, originality, and diversity, removes unsupported claims and duplicate coverage, and produces one newsletter of approximately 600–750 words.</p><p>The objective is not to combine four drafts blindly. It is to use different model perspectives to create one useful newsletter.</p><p><em>Here is the multi-agent architecture at a glance:</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BFpSJdN0XS9urPIbaWRjGw.png" /></figure><h3>5. Designing for Failures and Delays</h3><p>During my first multi-agent run, Google’s API returned a 503 error. Fortunately, the other agents continued working.</p><p>I established a simple rule: <strong>if at least two of the four writers succeed, the newsletter can proceed.</strong></p><p>I also added retries for temporary provider failures and an eight-minute deadline for the writer stage. If fewer than two writers succeed, the content is insufficient, or the editor fails, the workflow stops without sending the newsletter.</p><p>The editor remains a single point of failure, so a backup editor is one of my next improvements.</p><h3>6. Adding Memory and Improving Email Design</h3><p>A daily newsletter should not repeat yesterday’s stories or send duplicate emails.</p><p>I introduced a JSON history file that stores previously delivered issues, headlines, timestamps, and source links. Writers and the editor use this history to avoid repeated stories, while application code performs additional duplicate checks.</p><p>I also implemented a 12-hour minimum gap between deliveries. This prevents accidental duplicate runs and avoids unnecessary API costs. Since GitHub Actions starts fresh for each run, the history is carried between executions using its cache.</p><p>On the presentation side, I redesigned the email to resemble a clean technology publication, with clear headings, readable typography, source links, and an estimated reading time.</p><p>The --resend option lets me preview layout changes using an existing newsletter without making additional AI calls.</p><h3>7. What I Learned</h3><p>Building this project reinforced several important engineering principles:</p><ul><li><strong>Let AI generate; let code enforce.</strong> Critical rules such as source validation, word count, duplicate detection, and delivery limits should be checked by application code.</li><li><strong>Give agents capabilities, not just instructions.</strong> Web search made the difference between generating plausible news and researching current stories.</li><li><strong>Expect partial failures.</strong> Multiple providers improve diversity, but retries, timeouts, and fallback rules are essential.</li><li><strong>Make failures visible.</strong> A broken workflow should alert me; an intentionally skipped run should exit quietly.</li><li><strong>Keep development inexpensive.</strong> Dry runs, cached content, and automated tests make experimentation faster. My 93 automated tests run in approximately one second without real API calls or email delivery. Built and tested for under $10, using low-cost models and free tiers wherever possible. It can also run weekly instead of daily to cut costs further.</li></ul><p><em>Here’s a sample newsletter format for reference.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x_zzvXottCY80AVl-AK1AQ.jpeg" /></figure><h3>Final thoughts</h3><p>This project started with a single prompt and gradually evolved into a system that coordinates multiple AI providers, validates content, handles failures, maintains history, and delivers a newsletter automatically.</p><p><strong>My biggest takeaway: building a reliable AI agent is not just about choosing a model or writing a good prompt. The real engineering work lies in the tools, integrations, validation, memory, and failure handling that make the system dependable.</strong></p><p>If you are building your own AI agent, start small. Get the integrations right first, then add capabilities and reliability controls step by step.</p><p>The code and documentation are available on GitHub: <a href="https://github.com/Vijayavel/MultiAgentNewsletter">https://github.com/Vijayavel/MultiAgentNewsletter</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6e20c3b8626d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[5 Podcasts that Increase your Software skills]]></title>
            <link>https://medium.com/@vijayavel.p/5-podcasts-that-increase-your-software-skills-5b0e4853a4b4?source=rss-33350d5fc5f2------2</link>
            <guid isPermaLink="false">https://medium.com/p/5b0e4853a4b4</guid>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[learning-to-code]]></category>
            <category><![CDATA[listening]]></category>
            <category><![CDATA[podcast]]></category>
            <dc:creator><![CDATA[Vijayavel Palaniappan]]></dc:creator>
            <pubDate>Fri, 05 Jun 2020 14:03:36 GMT</pubDate>
            <atom:updated>2020-06-05T14:03:36.259Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VNUXHI8Eo-6arVU0_0lsZg.jpeg" /></figure><p>Listening to podcasts is one of the easiest ways to increase knowledge while driving, traveling or relaxing time. I have listed out the top 5 Podcasts available based on their popularity and variety.</p><p><a href="https://softwareengineeringdaily.com/"><strong>1.Software Engineering Daily</strong></a></p><p>It is a San Francisco Bay area based Podcast having a variety of topics, mainly focused on technical ideas, technology behind solving business problem ranging from Startup companies to big software giants. It is an hour interview format series hosted by Jeff Meyerson. He always sets the context to the topic in a way that everyone can understand and get most out of it. It also has a series of Podcasts related to Cyber Security, UI frameworks etc. It is one of the most active Podcasts focusing on up to date topics.</p><p><a href="https://www.codingblocks.net/"><strong>2. Coding Blocks</strong></a></p><p>Three professional developers namely Allen Underwood, Joe Zack and Michael Outlaw discuss about variety of Software topics current in the market. They have sections like survey, tip of the week and a core topic. Meeting notes of the topic, survey results and tip of week is available in the website. Design Pattern series, book review of Clean Architecture and Designing Data intensive applications are great discussion topics in their series. The duration of the podcast is little long and it goes beyond two hours, but you will enjoy it.</p><p><a href="https://spec.fm/podcasts/developer-tea"><strong>3. Developer Tea</strong></a></p><p>It is a motivational talk of 10 to 15 minutes for software engineers. It is clearly designed for the tea break or morning routine. Jonathan Cutrell delivers as average of 3 podcasts per week. The inspiring speech with software engineering flavor sets the day for better mind.</p><p><a href="https://changelog.com/"><strong>4. The ChangeLog</strong></a></p><p>The weekly shows about developer culture, software development, open source, building startups, artificial intelligence, and the people involved. The website has clear grouping of the categories. It is an interview style of audio stream lasting average of 75 minutes.</p><p><a href="https://hanselminutes.com/"><strong>5. HanselMinutes</strong></a></p><p>It is hosted by the famous Microsoft speaker Scott Hanselman. He covers subject ranging from new technologies, work culture, data science, agile, marketing etc. He interviews interesting personalities within the software world for 30 minutes. His questions are spot on and dive deep into the various topics.</p><p>The above podcasts are available in <a href="https://apps.apple.com/us/app/apple-podcasts/id525463029">iTunes Podcasts</a>, <a href="https://open.spotify.com/">Spotify</a> and other popular Podcasts applications. Remember to visit the respective podcast websites to get more information like transcripts, meeting notes and related blogs.</p><p>Enjoy listening!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5b0e4853a4b4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to get EST or anytime zones from .Net core micro-services]]></title>
            <link>https://medium.com/@vijayavel.p/how-to-get-est-or-anytime-zones-from-net-core-micro-services-a63915659dc0?source=rss-33350d5fc5f2------2</link>
            <guid isPermaLink="false">https://medium.com/p/a63915659dc0</guid>
            <category><![CDATA[dot-net-core]]></category>
            <category><![CDATA[microservices]]></category>
            <category><![CDATA[timezone]]></category>
            <category><![CDATA[operating-systems]]></category>
            <category><![CDATA[cloud-services]]></category>
            <dc:creator><![CDATA[Vijayavel Palaniappan]]></dc:creator>
            <pubDate>Tue, 04 Feb 2020 07:03:59 GMT</pubDate>
            <atom:updated>2020-02-04T07:03:59.234Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>Getting local-time of specific Time Zone using .Net Core micro-services in Cloud</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gs4fLhRYedUy9IlA-k647g.jpeg" /></figure><p>We normally come across business use cases to convert one time zone to another time zones or get local time of a particular city/region. For example, many banking or trading applications are globally used in various time zones. Business rules like transaction cut-off times, market open/close times are bound to a particular time-zone. In many cases, services handle business logic to control time bound rules for these kind of applications.</p><p>In the modern cloud software architecture, micro-services are hosted in different operating systems and various data centers across the world. When .Net Core micro-services hosted in any cloud platform such as AWS, Azure, Google Cloud etc, the below discussed solution gives the correct time of the given time-zone irrespective of their data center locations. It handles most commonly used operating systems namely Windows, Linux and Mac OS in cloud architecture in all time zones.</p><p>.Net Core has <em>RuntimeInformation</em> class in <em>System.Runtime.InteropServices</em> library. It provides operating system details which helps to find the right string of the time zone. In this example, it helps to find the eastern time zone. This class is supported in all versions of .Net Core and .Net framework.</p><p>Another class <em>TimeZoneInfo</em> is used to convert one time zone to another. Here server machine gives the UTC time and it is converted into Eastern time.</p><pre>public static DateTime GetEasternTimeZone()<br>{<br>    TimeZoneInfo easternTimeZone = null;<br>    if (RuntimeInformation.IsOSPlatform(OSPlatform.Windows))<br>    {<br>       easternTimeZone = TimeZoneInfo<br>           .FindSystemTimeZoneById(&quot;Eastern Standard Time&quot;);<br>    }<br>    else if (RuntimeInformation.IsOSPlatform(OSPlatform.Linux) <br>                || RuntimeInformation.IsOSPlatform(OSPlatform.OSX))<br>    {<br>        easternTimeZone = TimeZoneInfo<br>            .FindSystemTimeZoneById(&quot;America/New_York&quot;);<br>    }<br>    else<br>    {<br>        throw new ArgumentException(&quot;Not supported OS&quot;);<br>    }<br>    return TimeZoneInfo<br>              .ConvertTimeFromUtc(DateTime.UtcNow, easternTimeZone);<br> }</pre><p>Time-zone strings vary region to region, it needs to be configured in the application code. Normally time zone string vary one operating system to another. In this code, I have given an example of Eastern/New York time zone for all major operating systems. Similarly, you can configure any region in the .Net Core micro-service.</p><p>The list of time zone strings for Windows OS compatible found <a href="https://support.microsoft.com/en-us/help/973627/microsoft-time-zone-index-values">here</a>.</p><p>For Mac OS you can find the time zone strings using command</p><pre>sudo systemsetup -listtimezones</pre><p>For Linux, time zone names can be found by <em>zoneinfo</em> and drill down of zones gives specific city or region.</p><pre>$ ls /usr/share/zoneinfo/</pre><p>I have used the above approach in one of the cloud services.</p><p>There are alternate options available in the form of external/third party APIs, if your use-case allows to consume external APIs, it is worth considering <a href="http://worldtimeapi.org/">http://worldtimeapi.org/</a>, it is free version of simple rest API.</p><p>Thank you for reading.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a63915659dc0" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>