Skip to main content

Digital Discipline 2.0: Mesh Networks, NextDNS, and the Ongoing Battle Against YouTube

· 3 min read
D Balaji
Lead Design Technologist

Quick Recap

We took our first swing at digital hygiene: NextDNS configured on a non-ISP router, mobile Private DNS set, the whole house subtly steering away from distractions. Our battle star? The never-ending chase to block YouTube.com for good.

But as always, the story gets more interesting.

The Parental Control Problem — ISP Routers Just Don’t Cut It

The so-called “parental controls” on most ISP-supplied hardware are a masquerade. The DNS fields are locked down; your hands are tied. For real filtering, the only way out is to move to your own hardware so I picked up a TP-Link mesh router supporting true mesh networking.

Mesh brings coverage, flexibility, and—crucially—three distinct SSIDs:

  • Default
  • Guest
  • Smart Devices

Now, with NextDNS set at the router level, everything in the house bows to centralized rules.

NextDNS Profiles: Access Levels Like IAM for Humans

One genius detail about NextDNS is its profile-based filtering, think of it like authorization groups in IAM. Here’s how my rules stack up:

  • Default: YouTube is blocked everywhere, the good citizens’ network.
  • Elders: YouTube allowed (but only when the baby’s not around).
  • Dev Mode: Everything permitted—for me, when it’s time to break things or fix them.

Side effect: some analytics break, and anyone with the Tor app can waltz past the controls. It’s not bulletproof, but it’s pretty resilient.

SaaS Fatigue and the DIY Fix

Here’s the catch: NextDNS’s free plan caps at 30,000 DNS requests/month. That’s enough for one or two users, but add a full house and you’ll hit the wall. The fear: getting locked into yet another SaaS subscription for every single filter.

  • Temporary solution: Careful rationing, multiple profiles, and possibly automating some extra DNS protection. Still, I have DIY plans brewing to avoid lifelong lock-in.

The Result: Under Control… For Now

With distinct SSIDs, granular DNS filtering, and profile-based controls, the home network feels far more intentional and secure. The only wildcard? Human factors, will elders and guests play along or look for loopholes?

I personally save nearly 2 hours a day with no youtube, thats 30 days saved per year.

But for now, digital discipline is stronger than ever. Parenthood is chaos; your network doesn’t have to be.


To fellow parents: tweaking your network takes patience, but the peace of mind is worth it. On to the next frontier…

[1] https://dhbalaji.dev/blog/2025/preparing-home-network-before-first-child

From Tragedy to Triumph: My New Laptop Specs

· 4 min read
D Balaji
Lead Design Technologist

It began as most heartbreaks do, not with a warning, but with a moment. Eight years of companionship, memories etched into plastic keys and flickering screens. This was not just a laptop; it was a cherished confidant, a silent witness to late-night hustles, travels, and the fragile trust between creator and machine.

But even strong bonds can snap. On a monsoon-drenched evening, fate delivered a blow. The Dell service center greeted me somberly, where a sticker cheerily warned about “monsoon laptop care”—advice, in my case, a little too late. The diagnosis: my beloved Dell Inspiron, dead on arrival. There were no heroic rescues, just a silent mourning and the slow ache that follows a familiar loss.

The fallout was immediate. Backlogs grew like weeds. Blogs languished in drafts as an interim Dell Vostro—salvaged from the garage and coaxed to life with Debian—struggled beneath the weight of creative ambition. The void left by the Inspiron haunted every click. Sooner or later, I knew: it was time to let go and move on.

The Pursuit of Power And Value

I dreamt, briefly, of a desktop—a machine built on ChatGPT-powered advice, futuristic and mighty. But reality grinned: a 3–5 lakh rupee price tag. It felt absurd, the cost outweighing the gains, with the grey cloud of warranty uncertainty looming. Dell’s tower PCs were impossible to “add to cart” without wading through conversations with bots; the tempting A51 desktop was a crush at first sight, quickly doomed by its price—well above ₹5 lakhs.

The configurations under Dell’s “Workstation” banner cost more than a new Ather Pro scooter. Disheartened, I wandered into the wilds of OLX, mentorless but hopeful.

On OLX, the market teemed with budget laptops—plentiful at 25k, but the beasts I needed lurked rarely, cloaked behind search filters: “Dell i7”, “50k and above”. And yet, despite Apple’s siren song and options aplenty, my heart stayed loyal to Dell—perhaps from habit, perhaps nostalgia.

On OLX, two sellers emerged: individuals hoping for a fair price, and seasoned shopkeepers flipping out-of-warranty stock. Caution prevailed; I chose the latter. Stranger meets in unknown places—money flashing on UPI—felt like a recipe for disaster. I wanted reliability, not adventure.

The Resurrection: Enter Dell 7620

RG Computer, Belandur was unassuming—a small shop, a friendly shopkeeper, and an attentive assistant. There, among reclaimed silicon, sat my contender: the Dell 7560. (Specs to come.) The inspection ritual began:

  • Checked the Dell service tag, verified the model’s lineage.
  • Scanned for scars—deep or faint—from previous lives.
  • BIOS revealed battery health and year of manufacture (both scrutinized).
  • Booted and rebooted, measuring patience.
  • Ran HW Info, checked SSD specs—ensuring no deceit hid beneath the surface.

The shopkeeper walked me through their refurbishing process—old NVMe swapped for a fresh NVMe, a subtle upsell attempt gracefully declined. After two hours, I still forgot to check every key, but luck favored me; all were functional.

Armed with a bill and three months of shop-backed warranty, I walked out—lighter in heart, heavier in hand. The workstation is, in a word, solid. I do not recommend carrying it around on your back, its indeed heavy. I use my Dell Vostro for light work on the move.

Meet the Beast: Dell 7620

  • Intel i7, 11th Gen, 8 cores @2.50GHz
  • 32GB DDR4 RAM
  • 512GB healthy NVMe storage
  • Nvidia 6GB graphics
  • Windows 11 Pro

Things I Don’t Understand Yet

  • The 180W charger—why such a power-hungry beast for this machine?
  • The air vents, reminiscent of a car radiator, puzzling me with their design and purpose.
  • The CPU core temperatures, often hovering at 50°C and above—stable or stealth warning?

Red Flags? Of course:

  • Out of Dell warranty, over four years old
  • Noticeable signs of prior use
  • Fingerprint sensor, more gimmick than guardian

But for ₹55k? In my urgency, the deal felt fair. Back at the desk, it stood firm—two USB-Cs, two USBs, HDMI, and a DP port. It powers two monitors, a cooling pad, and more. The webcam is decent (bonus: a privacy slider).

Epilogue: Grief, Growth, Gratitude

Three weeks in, I’m surprised by the stability. I’m building again—products, content, momentum. The wound from the monsoon disaster remains, but it’s fading, replaced by a story of resilience and renewal.

Overall rating: 4.5/5. For a refurbished warrior and a rekindled creator’s journey.

Baby on Board, Firewalls Up, A New Dad’s Guide to Internet Hygiene at Home

· 5 min read
D Balaji
Lead Design Technologist

We’re expecting a baby soon, and as I prepare for sleepless nights and diaper disasters, there’s one more thing I want to get ahead of — the internet. Yes, the same magical place that helps make money is also crawling with algorithmic predators waiting to turn our kids into scroll zombies before they even learn to walk.

So here’s my no-fluff, nerd-dad guide to securing your digital den — whether you're tech-savvy or just tech-suspecting.


Step 1: Understand Your Home Network Like a Responsible Adult

Most homes today are multi-network setups, even if they don’t realize it.

Count the Networks

Total Networks at home = 1 broadband + each phone with an active mobile data plan

That means your child's screen time could sneak in through any of these entry points. First step: identify and isolate them.

Basic Network Topology (Before Changes)

Internet → ISP Router (Wi-Fi & LAN) → All Devices → your child

This gives the ISP's router full control over DNS, firewall rules, and device access. Not ideal for a security-conscious parent.

Smart phones + data packs → your child

Data packs by carrier dont give any parental control tools.

We need to target the network level filtering which is generally good enough to block ads, parked domains and time wasters.


Step 2: Fix the Manhole — Lock Down Your Broadband

Why ISP Routers Are a Problem

  • Most ISP routers are locked down — DNS settings greyed out
  • Firmware is often outdated and insecure
  • Limited or no parental control features
  • No custom firewall or VLAN options

What I Did Instead

  1. Disabled Wi-Fi on the ISP router
    Let it act as a basic modem or Layer 2 switch.

  2. Introduced My Own Router (Netgear)
    Plugged into the ISP box via Ethernet (LAN to WAN). This separates my internal LAN from the ISP's network.

  3. Enabled NAT, DHCP, and DNS on My Router
    Configured to issue private IPs (e.g., 192.168.1.x) and point all DNS queries to OpenDNS servers which look like:

    • 208.67.222.222
    • 208.67.220.220
  4. Registered with OpenDNS
    Created a free account, added my dynamic IP, and enabled category-level filtering. You can block:

    • Social networks
    • Video sharing
    • Adult content
    • Gaming, shopping, forums — up to you
  5. Tested the Setup
    Verified DNS override by visiting:
    https://welcome.opendns.com

Updated Network Topology

All devices now talk to the internet via my firewall + DNS filters, not the ISP’s.


Step 3: Taming the Phones — Especially the Ones With Data Packs

Broadband is fixed, but what about mobile data on phones? These are the real backdoors.

Simple But Effective Plan

  • Recharge only when needed (no monthly auto-renewal of data-heavy packs)
  • Use mobile data only for OTPs, emergency maps, and short bursts

The Engineer's Setup

I needed something robust and centrally managed.

  1. For Broadband Devices – I use OpenDNS via the router.
  2. For Mobile Data (4G/5G) Devices – I use NextDNS via Android/iPhone's Private DNS feature.

Why NextDNS?

  • Supports DNS-over-TLS (DoT) and DNS-over-HTTPS (DoH), apps I control you.
  • Per-device logs, analytics, and block rules
  • Works even over mobile data, friends hotspot
  • One profile can be shared across multiple devices
  • For entire family, basic plan is worth it. I dont think free plan will make it.

How to Set It Up on Android

  • Go to: Settings > Network & Internet > Private DNS
  • Choose: Private DNS provider hostname
  • Enter: your-profile-id.dns.nextdns.io (from nextdns.io dashboard)
  • Save and test at: https://test.nextdns.io

Now even on 4G, your DNS traffic is encrypted and filtered.


The Bigger Motive: Family-Wide Digital Discipline

This isn’t just about protecting a newborn. It’s about reprogramming how we use the internet — so that the baby doesn’t see everyone glued to screens 24/7.

What We’re Trying to Do

  • Remove addictive triggers from devices
  • Encourage meaningful use of tech
  • Create consistency across screens
  • Lead by example

This is not a one-man battle. It’s a full-stack family-level architecture decision.


Problems Faced

Technical

  • Changing DNS is straightforward if you’re used to router admin pages.
  • Most modern routers support this out of the box.

Psychological

  • Asking aging parents to give up binge-watching YouTube is like asking them to skip tea.
  • You’ll hear:
    • “What’s wrong with watching one or two reels?”
    • “It’s just a recipe video...”
  • Prepare for mini mutinies, emotional blackmail, and silent resistance.

Financial

  • Another subscription added, hopefully Netflix & Prime go away

But if you stick to it, it gets easier.


Advanced Nerd Notes

  • DNS Leak Protection: Use firewall rules to block outbound port 53 and redirect to OpenDNS.
  • Split VLANs: For homes with smart TVs and IoT devices, separate them into guest VLANs.
  • Device MAC Tracking: Some routers let you assign policies per MAC address (for kids vs adults).
  • Dynamic DNS: If your IP changes often, configure DDNS and bind it to OpenDNS/NextDNS updates.
  • Logs & Analytics: Use Pi-hole or Netflow on a Raspberry Pi to monitor domain requests for forensic insights.

Disclaimer

I’m not selling any tool here. This blog is about the mindset shift required to raise children in the age of digital dopamine.

Use any tool you prefer. The idea is to build a secure, distraction-minimal environment where children grow up watching their parents being intentional — not just entertained.


Final Words from a Nerdy New Dad

Failing to plan is planning to fail. We’ve baby-proofed our furniture. Now let’s baby-proof our Wi-Fi.

Because the real "parental control" isn’t in a router setting — it’s the decision to take control in the first place. Also do let me know what are your strategies? Did you deny lending your phone?

Trifles You’ll Spot in Every Tech Team Even In The Age of AI

· 5 min read
D Balaji
Lead Design Technologist

Sometimes you just need to break the writer’s slump, and what better way than to air some tech community laundry/gossips. Here are 3 classic struggles that keep popping up, no matter how fast the AI world moves.


1. Senior vs. Junior Conundrum – The Banyan Tree Effect

Juniors often feel like they’re growing in the shadow of their seniors – those strong, wide-spread banyan trees that block opportunities and stunt their growth. This isn’t just a theory – data shows that nearly 30% of techies with less than 3 years of experience switch jobs, looking for more sunlight and fresh air.

For juniors, the challenge isn’t just the learning curve but finding their voice in a team where the seniors have already taken the best seats. It’s like trying to grow a startup in a market dominated by FAANG.

While seniors are busy debating the best design pattern for a microservice, juniors are trying to get their first pull request merged without being roasted on the code review thread. It’s a constant game of "How many comments will this PR get before it’s finally approved?", "Will anyone mentor me honestly about all corners of the long career ahead?", etc.

Seniors, on the other hand, see juniors as a potential threat – the fresh minds who come in armed with JavaScript, Python, and React and casually throw around terms like LLMs, prompt engineering, hackathons and vector databases. Meanwhile, the seniors are still perfecting the art of Spring Boot and JPA, occasionally grumbling about how Java 8 was the last “real” upgrade.

It’s not just about skills – it’s about mindset. The seniors have survived production outages at 3 AM, managed political office battles, and honestly mentored a generation of developers who now run startups.

Juniors, on the other hand, are still fresh, living out of PG rooms, debugging their code on 15-inch laptops, and pushing code on a 5G hotspot while dodging their roommate’s PUBG screams. They may lack experience, but they have the raw hustle, curiosity, and caffeine tolerance that seniors often lose along the way.

And if you want a real generational gap, just bring up Agile vs. Waterfall in a meeting. The seniors will reminisce about the days when project plans were thicker than the SRS document, while the juniors wonder if they should include Agile Scrum in their LinkedIn profile.


2. Employee vs. Management – The Cost-Cutting Showdown

When it comes to cutting costs, management often takes the path of least resistance – reduce headcount. Performance, age, location, experience – all fair game when trimming the payroll. This is the "high-impact, low-cost" strategy that features in every MBA case study.

From the management perspective, it’s all about "optimizing the org chart" and "maximizing shareholder value" – fancy phrases that essentially mean fewer people, more profit. The logic is simple: cutting the bottom 10% might save the company millions without the PR disaster of reducing executive perks.

But employees see it differently. To them, the real savings lie in cutting the fat at the top. After all, the CEO earning 100x the average developer’s salary, VPs flying business class, and middle managers enjoying plush perks attending meetings.

And then there’s the perks debate. Employees argue that cutting free food, and fancy offices in favor of WFH (Work From Home) could save millions. After all, those ping-pong tables and bean bags aren’t exactly mission-critical. And let’s be real – the only people who truly use the office gym are the same ones who have the time for marathon LinkedIn posts.

This is especially stark in India, where C-level executives at top tech firms earn crores per annum, while a fresh developer might make 7-15 LPA. The disparity is not just in the paychecks but also in the power to decide the fate of others.


3. The Outsourcing Squeeze – A Tale of Two Realities

Outsourcing is a simple idea – move work to a cheaper location and save big. India, with its massive pool of highly skilled engineers, is a top destination for this. In 2024 alone, India’s IT exports were estimated at over $250 billion, with GCCs (Global Capability Centers) in Bangalore, Hyderabad, and Chennai leading the charge.

But this efficiency comes with a human cost. For every American worker laid off, there’s likely a replacement in Bangalore, Manila, or Chengdu doing the same work at a fraction of the cost. It’s the "rob Peter to pay Paul" strategy of the corporate world.

But it’s not a bed of roses for the outsourcers either. They’re often stuck in corporate jungles – the GCCs of India – doing the same work for 3x-5x less pay than their Western counterparts, with fewer perks, less vacation, and the constant contract renewal threats.

And to add to their woes, the entry-level salaries at these GCCs have barely moved in the past 15 years, despite inflation and the rising cost of living. It’s like getting a static variable in a dynamic world.

Plus, they have to sync up at 11 PM for calls with PST time zone clients who think that IST stands for "I’ll Slack Tomorrow."


This post is a simple rant for the slump – not to be taken seriously. There are many more battles in tech, like open source vs. closed source, cloud vs. on-prem, and Windows vs. Mac. I’ll tackle those when the next writer’s block hits. 😉

Bridging the Gap: CAP Theorem for Senior React Developers

· 5 min read
D Balaji
Lead Design Technologist

Why this post? As frontend engineers, we often focus narrowly on frameworks and tooling—primarily JavaScript, React, and UI libraries. But many of us hit a career plateau because we lack exposure to core software engineering principles.

This post is part of a growing genre I call “Bridge Posts”—connecting frontend development to foundational software architecture concepts. The goal is to help frontend engineers think like system designers, not just component builders.

Today, we explore the CAP Theorem, a classic principle in distributed systems, and map it to familiar frontend scenarios—such as Git workflows, real-time collaborative UIs, and offline-friendly apps.


Understanding the CAP Theorem

In distributed systems, the CAP Theorem states that during a network partition (i.e., some parts of the system can’t communicate), you can only guarantee two of the following three:

  • Consistency (C): All nodes see the same data at the same time.
  • Availability (A): Every request receives a response—regardless of the freshness of the data.
  • Partition Tolerance (P): The system continues to operate even when parts of it can’t communicate.

In practice, partition tolerance is non-negotiable in any distributed system. Therefore, systems must choose between consistency and availability when partitions occur.


Git Analogy: You Already Use CAP

Let’s start with Git—a tool every developer knows.

  • When you commit locally on a plane, you’re in a partitioned state.
  • You can continue working (Availability), even though your code may diverge from your teammate’s (Consistency is compromised).
  • Once reconnected, you merge changes to restore consistency.

Git is an AP (Available + Partition-Tolerant) system. It tolerates partitions and lets you work offline but eventually requires reconciliation.


React Use Case: Real-Time Collaborative Forms

Now imagine you’re building a collaborative form in React. Multiple users edit the same form in real-time. Updates are synchronized via WebSockets or polling.

How does CAP play out here?

1. CP – Consistent & Partition-Tolerant

  • If a user loses network connectivity, editing is disabled.
  • This ensures everyone always sees the latest state.
  • However, the application becomes unavailable for offline or disconnected users.

Use cases: Healthcare apps, finance platforms—where data integrity is paramount.

2. AP – Available & Partition-Tolerant

  • Users can continue editing offline.
  • Changes are stored locally and synced later.
  • This may lead to conflicting edits, requiring merge strategies.

Use cases: Note-taking apps, chat applications—where user flow matters more than perfect sync.

3. CA – Consistent & Available (No Partition Tolerance)

  • Works as expected under perfect network conditions.
  • Any partition causes the system to fail or block.
  • While theoretically ideal, this model is impractical in real-world distributed systems.

🌐 Designing for Partition Tolerance

A network partition occurs when different components of a distributed system—clients, services, or databases—cannot communicate due to a temporary network failure. Each component may still be operational, but they're isolated like islands without bridges.

In frontend development, this is surprisingly common:

  • A user loses internet connectivity mid-session.
  • A mobile app hits a dead spot with no signal.
  • The frontend can reach a CDN or cache but not the main API server.

Designing for Partition Tolerance means your app should continue functioning as gracefully as possible, even during such disconnects.

As a React developer, this involves:

  • Storing user actions locally (memory, localStorage, IndexedDB).
  • Queuing mutations and syncing later (e.g., Service Workers, Apollo cache, Redux middleware).
  • Providing clear UI cues: “You’re offline, changes will sync later.”
  • Implementing conflict resolution logic, if needed.

Real-world examples:

  • Figma continues rendering and recording user edits during disconnects.
  • Notion lets you type offline and syncs the block tree later.
  • Gmail stores draft emails offline and sends them once reconnected.

These applications opt for Partition Tolerance, ensuring the app remains usable—even if consistency is delayed or temporarily broken.

Designing for Partition Tolerance doesn’t mean ignoring consistency—it means accepting that consistency might be eventual, not immediate.

In distributed systems, network failures are not rare edge cases—they're expected events. As frontend engineers, acknowledging and designing for them elevates your thinking from component trees to system-level resilience.


Mapping CAP to Frontend Patterns

Frontend PatternCAP TradeoffNotes
React Query (stale-while-revalidate)APShows stale cache first, fetches fresh data.
Optimistic UI (e.g., message send)APAssumes success and syncs with the server later.
Disabling forms on lost connectionCPPrevents stale writes by enforcing consistency.
Service Workers / Offline-First PWAAPOperates offline and reconciles post-reconnect.
Live collaboration (e.g., Figma, Google Docs)AP + conflict resolutionResolves sync issues with operational transforms.

What This Means for Frontend Developers

You don’t have to be building a distributed database to care about CAP. If your application:

  • Caches remote data
  • Lets users work offline
  • Supports multi-user collaboration
  • Relies on eventual consistency

…then you’re actively navigating CAP trade-offs.

Ask yourself:

  • Can users work with stale data? → Choose Availability.
  • Must every write be accurate and conflict-free? → Prioritize Consistency.
  • Should the app always respond—even during outages? → Design for Partition Tolerance.

Closing Thoughts: Frontend as a Distributed System

“CAP isn’t just a backend concern—it manifests in every interactive, networked UI you build.”

Whether you’re building a rich client with React Query, crafting optimistic updates, or designing for offline-first usage, you’re constantly making trade-offs. Understanding CAP helps you make them consciously.

This post was part of a broader mission to elevate frontend engineers into system thinkers—developers who don’t just build buttons, but design resilient user experiences.

Let’s not be cookie-cutter React developers. Let’s bridge the gap.


Would you like me to help you turn this into a blog-ready Markdown + SEO-friendly format?

Back to stage, My Comeback to Toastmasters and First Visit to Hosur Toastmasters Club

· 4 min read
D Balaji
Lead Design Technologist

What happens when a frontend architect meets a podium after years? You get a mix of console.log("confidence") and some seriously good speeches.

I'm dhbalaji — Lead Frontend Engineer by day, Toastmaster since 2012, and collector of certificates like CC, CL, and ACB. After taking a break to focus to master sorcerous Javascript. But when I heard there’s a Toastmasters Club right here in Hosur, my curiosity wanted me to do a little mic check.

So I walked in. Here's what happened.


Hosur Toastmasters Meeting Invite

📍 Where It Happened

  • When: Every Sunday, 10:00 AM – 11:30 AM
  • Where: 2nd Floor, Finance Academy, ASTC HUDCO, Hosur
  • Parking: Loads of space — park like a boss
  • Setup: Cosy room with a ~30 person capacity
  • Hybrid Mode: Online guests can join via Microsoft Teams

🗺️ If you're Googling "Toastmasters club near me in Hosur" — this is it! Hosur Toastmasters club permalink


👀 First Impressions

Having attended clubs with bells, whistles, and near-TED levels of polish, Hosur Toastmasters felt refreshingly raw. Think: garage band with talent, waiting to hit Spotify.

✅ What Worked:

  • Diverse Professionals: Engineers, finance folks, HR pros — all under one roof
  • Mature Audience: 30s+ crowd added depth and real-life stories
  • Warm Vibes: Received a formal invite ahead of time (yay structure!)

⚠️ What Could Be Better:

  • Sunday morning? Great for early birds, rough for Netflix-bingers
  • Limited networking: Hybrid format made hallway chats tough
  • Retention Blues: Noticed some churn — fresh energy needed

The session I attended was a joint meeting with Proficient Toastmasters Club, with quite a few tuning in online. Felt like an office teams meeting — but with better grammar.


🌟 Meeting Highlights

I won’t pretend I was the General Evaluator, but here’s what stuck with me:

  • Theme: “Voices of Nature” — poetic, calming, not a weather report
  • Word of the Day:Cacophony” — ironically, the meeting room fan was anything but
  • Prepared Speeches: Delivered with confidence — definitely mentor material
  • Table Topics: Real-time adrenaline for guests. I spoke. I survived.
  • TAG Team: Tracked Time, Ah-counter, and Grammar like ninjas with excel sheets, yes someone ended up sharing screen on teams
  • Evaluations & Awards: Encouragement with a touch of pageantry and photos
  • Post-Meeting Chats: Surprisingly rich — could have been better

🤔 What Could Level-Up the Experience

Here’s my candid guest audit (no PowerPoint, promise):

  • Introduce guests early — helps with context and connection
  • Bigger in-person crowd = more energy
  • Meeting decorum — a little polish goes a long way
  • Tighter meeting flow — less lag between segments
  • Record & upload speeches — boost visibility and speaker growth

💬 People Who Inspired Me

This was the deal-sealer.

Where else do frontend developers, sales pros, finance strategists, and HR veterans talk about communication, confidence, and community — all in the same room?

Special shoutout to TM Muthu Kalimuthu, the Club President. His hospitality and thoughtful questions before the meeting made me feel like I was already part of the team.


🧠 Takeaways from Hosur Toastmasters Club

  • Hosur isn’t just about factories anymore. We’ve got IIT training centres, shopping malls... and now, Toastmasters.
  • The club may not be in full bloom yet, but the seeds are solid. Leadership can make it shine.
  • English fluency can be a barrier in this region — this club is a bridge, not just a stage.

👍 Should You Join Hosur Toastmasters?

Yes, yes, a thousand times yes — especially if you:

  • Want to improve your public speaking and leadership skills
  • Are looking for community engagement in Hosur
  • Enjoy learning from people outside your industry bubble

Here’s how I imagine the club’s golden mix:

  • 33% – Students and early-career folks
  • 33% – Regulars actively participating and giving speeches
  • 33% – Mentors and seasoned Toastmasters offering wisdom from the backbenches

😆 Fun Goof of the Day

Speaker (with full confidence): “And then I heard the birds slurping...” instead of "birds chirrping"

Audience: blink blink Grammarian: Noted


💬 Final Thoughts

I didn’t just walk into a meeting. I walked into a reboot.

So if you’re searching for:

  • Hosur Toastmasters Club timings
  • Toastmasters near ASTC HUDCO
  • How to improve public speaking in Hosur

You’ve found your place.

🎤 Come for the speeches. Stay for the people. Rediscover your voice.

Understanding How LLMs Work with a Doctor's Clinic Analogy

· 5 min read
D Balaji
Lead Design Technologist

Large Language Models (LLMs) are revolutionizing the way we build intelligent applications—especially in frontend development. From ChatGPT to custom AI copilots, LLMs are everywhere. But have you ever wondered how LLMs actually work under the hood?

The theory behind LLMs can be complex—terms like embeddings, transformers, tokenization, and vector space often feel overwhelming. That’s why I’m using a familiar analogy: a visit to a local doctor’s clinic.

This blog post breaks down the internal architecture of LLMs using a real-world story that’s easy to visualize and remember. Whether you're a developer exploring AI integration, an engineer curious about embeddings, or someone building chat interfaces, this analogy-driven explanation will help you understand:

  • What tokenization really means
  • Why vectors and embeddings are essential
  • How transformers "think" using attention
  • Why temperature affects LLM creativity
  • How memory and context shape outputs

Let’s walk into the clinic. 🚶‍♂️💊


Characters in the Community Clinic Example

  • User – A person interacting with the LLM (the patient)
  • LLM – The large language model (the doctor)
  • Internals – Sequence of steps the model uses to generate a response (diagnostic workflow)

Step 1: Patient Describes the Issue to Doctor

Scene:
The patient walks into the clinic.

Conversation:

  • Doctor: Hi, please sit down.
  • Patient: Thank you.
  • Doctor: Tell me.
  • Patient: I got back from my friend’s wedding 2 days back. Since yesterday, I feel cold and have body pains.
  • Doctor: (listening) hmm.

LLM Analogy:
This step is like tokenization — breaking the input sentence into smaller units (tokens) that the machine can understand.

Step 1 - Tokenizer:
Converts input from human language to machine-readable tokens.

  • Input: Plain text / story
  • Output: Word tokens (e.g., "wedding", "cold", "pains")

Step 2: Doctor Comprehends the Patient Problem

The doctor interprets the patient’s words and thinks in terms of temperature, symptoms, etc. They may even order diagnostic tests for more data.

LLM Analogy:
This is like the embedding layer, where word tokens are turned into number arrays (vectors) that hold semantic meaning.

Step 2 - Embedding Layer:
Converts tokens into vector form for semantic understanding.

  • Input: Word tokens
  • Output: Vectors that represent meaning (like symptoms turned into medical data)

Step 3: Doctor Runs Diagnosis Internally

The doctor thinks through the problem using experience and logic—checking symptoms against patterns they’ve seen before.

LLM Analogy:
This is the transformer architecture—especially self-attention layers, which compare words across the sentence to extract meaning and decide what matters most.

Step 3 - Transformer Layers:
Deep learning steps like self-attention and feedforward networks.

  • Input: Vectors
  • Output: Context-aware vectors based on internal learned patterns

Step 4: Doctor Considers Constraints (like budget)

Doctors don’t always prescribe expensive tests—patient affordability, practicality, and history all affect decisions.

LLM Analogy:
This step is like positional encoding—ensuring the model understands the order and structure of the sentence.

Step 4 - Positional Encoding:
Adds position-related meaning to vectors.

  • Input: Related vector tokens
  • Output: Ordered vector tokens based on sentence position

Step 5: Doctor Writes a Prescription

The doctor now documents the diagnosis and treatment. It’s a direct result of structured thinking and patient input.

LLM Analogy:
This is where the model decodes the internal vector into a predicted sequence of tokens.

Step 5 - Decoder:
Generates output tokens based on model’s confidence and logic.

  • Input: Positional vectors
  • Output: Probable tokens in correct context

Step 6: Doctor Shares the Prescription with the Patient

The final interaction. If the patient is nervous, doctor might tweak the recommendation. This is where human nuance enters.

LLM Analogy:
The LLM now converts tokens back into human-readable words. Here, temperature plays a role in how creative or safe the response is.

Step 6 - Output Layer:
Converts final vector into output text based on temperature setting.

  • Input: Final vector
  • Output: Natural language sentence

Bonus Step: The Doctor Remembers Your Previous Visit

LLMs with memory or context windows can remember prior conversations, like a doctor recognizing returning patients. This helps give better, contextual responses in multi-turn dialogues.


Summary

Just like a good doctor doesn’t Google your symptoms in front of you, a well-trained LLM doesn’t "think" in real-time. It applies complex math on pre-learned data to predict and autocomplete responses.

  • Words → Tokens → Vectors → Patterns → Tokens → Words
  • Everything happens in vector space, not in "language" as humans know it.

If you’re building with tools like LangChain, OpenAI, or embedding-powered RAG applications, understanding these LLM fundamentals gives you a huge advantage.

🧠 "A well-trained LLM is just a super-fast autocomplete that has read the entire internet."


✨ Takeaways for Developers

  • Tokenization = Parsing human input
  • Embedding = Understanding meaning
  • Transformer = Core logic engine / Deep learning
  • Positional Encoding = Sentence structure
  • Decoding = Constructing a response
  • Temperature = Tuning creativity
  • Context = Remembering past interactions

LLMs don’t reason like humans—but they recognize patterns with superhuman speed. Now you can too. If you like this story, share it on social media.

5 Things I am Letting Go of in 2025 - Building a Not-To-Do List

· 3 min read
D Balaji
Lead Design Technologist

As tech enthusiasts, we often talk about to-do lists to stay productive. But have you ever considered the power of a not-to-do list? It's about consciously identifying habits or activities that no longer serve your goals and removing them. Here's my not-to-do list for 2025, crafted with the intent to stay focused, efficient, and aligned with my evolving priorities.

1. No More Public GitHub Repos

Open-source contribution has been a cornerstone of the tech world, and I’ve had my share of excitement from it. But with the rise of AI tools, the landscape is changing. Unless you’re a core maintainer of a prominent library, the recognition that once came from showcasing code has diminished.

In 2025, I’ll shift my energy toward building meaningful products and functionalities rather than maintaining a public code showcase. The focus will be on solving problems and creating impact—where the real value lies.

2. No Social Media Overload

Social media can be a double-edged sword. It’s a goldmine for networking, but it’s also a time sink designed for immersion and doomscrolling. With AI tools providing quick insights (even without sitting through a 40-minute video), my reliance on these platforms will be minimal.

Here’s my plan: Engage with social media just once a week for an hour, in the following order:

  1. Facebook for personal updates.
  2. LinkedIn for professional networking.
  3. GitHub Feed to track industry trends.
  4. Twitter/X for niche topics and hot takes.
  5. YouTube Subscriptions for targeted learning.

Additionally, I’ll curate a list of websites to stay updated instead of depending on algorithm-driven feeds.

3. No Two Visits Per Week To Office

Working from the office puts you into a routine which is a supporting stick for those who lack self discipline of waking up early, switching subjects under work naturally. Thats why I am against working from office on select days of the week.

4. No More Reading Goals

I used to pride myself on being a voracious reader, devouring two books a week. But life evolves, and so do habits. For two consecutive years, I failed to hit my reading goals—and that’s okay.

In 2025, I’ll stop setting rigid reading goals. Instead, I’ll use mobile apps that provide daily book summaries to stay updated with the latest titles. Physical books will still have a place in my collection, but the pressure to meet arbitrary goals will be a thing of the past.

5. No More Video Courses Unless I’ve Tried and Failed

Video courses are everywhere—just one click away. But they often require hours of commitment, and many simply repackage information available in documentation or books.

Moving forward, I’ll treat video courses as a last resort. I’ll prioritize reading documentation, experimenting, and leveraging AI for quick answers. If I still hit roadblocks, only then will I invest time in a course. The same logic applies to interviews or tutorials—AI can summarize key insights faster than a 60-minute watch.


The Overall Theme: Build, Ship, and Upskill

2025 is about action over abstraction. Instead of endlessly documenting best practices or creating small libraries, I’ll focus on:

  • Building real applications for my portfolio.
  • Leveraging AI for 5x upskilling.
  • Prioritizing actual product development over pre/post-work content.

It’s time to let go of distractions and double down on meaningful work. Here’s to a focused and fulfilling year ahead!

What’s on your not-to-do list for 2025? Share in the comments—I’d love to hear your thoughts!