That is a supply-chain breach, not a direct hack of Vercel’s servers. But the effect is the same: my secrets were sitting in a system whose security depended on every vendor Vercel uses internally, and I had zero visibility into that chain.

For me, it was the third strike. Here is the full story.

Strike 1: Vercel’s per-seat pricing

I was happy on Vercel until I started collaborating.

I was running my projects on Vercel Pro with 4 collaborators. The bill: $160 per month. For a solo founder running multiple small projects with a few trusted contributors, that math stops making sense fast.

I did not want to gatekeep access to my own side projects to save money. So I moved to Netlify.

Strike 2: Netlify’s credit flip

Netlify used to have generous usage limits. Build minutes were metered clearly, and my fast builds fit comfortably in the lower tiers.

On September 5, 2025, Netlify moved all new accounts to credit-based pricing. Instead of build minutes being their own line item, everything (builds, bandwidth, function invocations, edge functions) draws from a unified credit pool.

What changed for me in practice:

• Builds got more expensive and harder to predict

• I was nudged onto the Personal plan to keep the lights on

• I ran into HTTPS/SSL issues on custom domains that took way too long to resolve

• Support was slow and felt scripted

None of this is a deal-breaker alone. Stacked together, it stopped feeling like infrastructure and started feeling like a relationship I had to manage.

Strike 3: The Vercel breach

Then the Vercel news dropped.

The part that stuck with me: even if Vercel’s engineers are world class (they are), my security posture was downstream of every SaaS vendor they use internally. A compromised laptop at Context.ai became my problem. I had no visibility into that chain, and no way to opt out.

The lesson was not “Vercel is bad.” The lesson was: platforms amplify supply-chain risk, and for projects where I care about who holds my secrets, that trade-off stopped making sense.

Why Hetzner?

Before getting to the stack, a quick word on provider choice. I looked at the usual suspects:

• DigitalOcean: Great UX, plenty of tutorials, but roughly 2x the price of Hetzner for equivalent specs

• Linode (now Akamai): Similar price to DigitalOcean, slightly more enterprise-oriented

• Vultr: Competitive pricing, more global regions than Hetzner, but less generous RAM per tier

Hetzner won for three reasons:

1. Best price-to-performance on the planet. A $6 Hetzner box matches a $12 DigitalOcean droplet on specs

2. EU data residency in Germany and Finland, which helps with GDPR if you have European users

3. Hetzner Cloud Backups at 20% of the server cost, cheaper than most competitors’ backup add-ons

The catch: only EU and US data centers. If you need low latency in Asia or South America, pair Hetzner with Cloudflare's CDN (I do this anyway, more on that below).

What I moved to

A $6 Hetzner VPS plus open-source tooling. Here is the full stack:

• Hetzner Cloud for the VPS

• Cloudflare in front of everything for DNS, CDN, and DDoS protection

• Coolify as my self-hosted PaaS layer (more on this below)

• Docker + Docker Compose for containerizing each project

• Caddy as reverse proxy. Handles automatic HTTPS via Let’s Encrypt, HTTP/2, HTTP/3, static file serving, with a config short enough to fit on a postcard

• UFW firewall with deny-by-default, only 22, 80, 443 open

• Fail2ban for brute-force protection on SSH

• Postgres in its own container with automated daily backups

• GitHub Actions pushing builds on merge to main

• Uptime Kuma for self-hosted monitoring

• Web3Forms for contact form submissions (more on email below)

Cloudflare: the free layer that should be a default

Before anything hits my VPS, traffic goes through Cloudflare’s free tier. That gets me:

• DDoS protection at the edge. My origin server never sees attack traffic

• Global CDN caching for static assets. Pages load fast in Asia and South America even though my server is in Germany

• Hidden origin IP. Attackers scanning the internet see Cloudflare, not my Hetzner box

• DNS management with instant propagation

• Free SSL at the edge (in addition to Caddy’s Let’s Encrypt certs at the origin)

This is not optional. If you run a VPS exposed to the internet, putting Cloudflare in front of it is table stakes.

Coolify: the piece that made this manageable

Coolify is a free, open-source, self-hosted alternative to Vercel, Netlify, and Heroku. You install it on your VPS, point your domain at it, and you get a clean web UI for:

• Git-push deploys. Connect a GitHub repo, Coolify builds and deploys automatically

• One-click services. Postgres, Redis, MongoDB, Wordpress, Plausible, anything in its template library spins up in under a minute

• Automatic HTTPS. Coolify handles Let’s Encrypt certificates for you, no Caddy config required if you do not want to touch it

• Built-in backups, logs, monitoring, and team access controls

• No per-seat pricing. Add as many collaborators as you want. Your only bill is the VPS

The trade: you own the box, so you still patch the OS and manage hardware-level backups. But the day-to-day deploy experience feels exactly like Vercel, without the invoice.

Email: the one thing I did not self-host

If you are wondering what SMTP I use, the honest answer is: I do not run my own mail server, and neither should you.

Self-hosting SMTP is a nightmare. Deliverability depends on IP reputation, SPF, DKIM, DMARC, reverse DNS, and the mood of whoever manages Gmail’s spam filter that week. Even seasoned sysadmins avoid it.

For the Echoscan site, contact form submissions go through Web3Forms, which is a simple form-to-email relay. The form posts to their API, they deliver it to my inbox. No backend code needed on my side.

For transactional emails on other projects (password resets, receipts, notifications), I use Resend or Postmark. Both are cheap, both have clean APIs, both handle deliverability for you.

Self-host what benefits from ownership. Outsource what benefits from network effects. Email deliverability is pure network effect.

Backups: the cheap way and the free way

Two honest options:

Paid and easy: Hetzner Cloud Backups. Hetzner charges 20% of the server base price for automated backups (7 daily snapshots rotated). On a $6 server, that is about $1 per month. Lowest-friction path.

Free and open source: Restic to Hetzner Storage Box. Restic is an encrypted, deduplicated backup tool. Point it at a Hetzner Storage Box or Backblaze B2, write a short cron job, done. Slightly more setup, more control, lower long-term cost as your data grows.

I’d pick the paid option unless you are managing many servers which I am not.

Hetzner snapshots: my “the box died” insurance policy

The obvious pushback to a single-VPS setup is: “what if the box dies?”

Fair question. My answer is Hetzner snapshots. A snapshot is a full point-in-time image of your server. With a fresh snapshot on hand, I can:

• Spin up a new Hetzner VPS from the snapshot in under 15 minutes

• Update the DNS record in Cloudflare

• Be back online before most people notice

Snapshots cost about $0.01 per GB per month. For a 40GB disk, that is roughly $0.40 a month. I take one before every big change and one on a weekly cron job.

Combined with daily automated backups, my recovery story is actually better than what I had on Vercel, where "recovery" meant redeploying and hoping the platform behaved.

The migration itself: zero downtime

Here is the part people worry about most, and it turned out to be the easiest.

Because everything ran in Docker locally, moving to the VPS was effectively a config copy plus a DNS switch:

1. Spun up the Hetzner VPS and installed Coolify

2. Deployed each project to the new VPS while the old Vercel/Netlify sites stayed live

3. Tested every route, every form, every database connection on the new setup using a temporary domain

4. When everything checked out, I flipped the DNS record in Cloudflare to point at the Hetzner IP

5. Done. No downtime, no broken links, no support tickets

That is the beauty of Docker. “It works on my machine” is finally just “it works.”

Trade-offs I won’t hide from you

This is not a free lunch. The honest ledger:

What you gain:

• Full control over your secrets, your stack, your data

• Predictable flat monthly cost

• Faster builds and deploys

• Real skills: Linux, networking, security hardening

• No per-seat tax as your team grows

What you take on:

• You are now the SRE. OS patches, security updates, log rotation, backups, all on you

• If the box dies at 2am, no one is paging an on-call engineer for you (although, I AM trying to automate this. More on this in the future)

• Zero-downtime deploys need a little thought (Coolify handles most of this)

• Initial setup might take a weekend if you are new to Linux.

For most small teams and indie founders, Coolify flattens that learning curve enough that the trade is worth it.

Who this move is for

Not everyone. Stay on Vercel or Netlify if:

• You are shipping fast and do not want to think about infrastructure

• You need global edge deployment with zero config

• Your team is large enough that per-seat pricing is a rounding error

Move to a VPS + Coolify if:

• You are running multiple small-to-medium projects

• Per-seat or credit pricing is starting to hurt

• You want to own your security posture end-to-end

• You are willing to invest a weekend to save serious money long term

The bigger point

Hosted platforms are convenient, and convenience has a price that is not always on the pricing page. Sometimes the price is per-seat. Sometimes it is credit volatility. Sometimes it is a supply-chain breach you did not sign up for.

Owning your stack is not the right answer for every project. But for small teams running their own products, the tooling has gotten good enough that “just run it yourself” is no longer the painful option it used to be.

I am not going back.

About Echoscan

This is what we do at Echoscan Software. We help small and mid-sized businesses:

• Set up cost-effective infrastructure like the stack I described here

• Automate repetitive processes with AI and RAG

• Bring down software costs by replacing overpriced SaaS with open-source alternatives you actually own

If your business is bleeding money on tools you do not need, or you want to modernize without handing your data to five different vendors, we can help.

Get in touch: echoscansoftware.com

In a recent interview of Sam Altman on Tree Hacks, he was asked the question -

“What AI subfield today feels like Open AI in 2016?”

“I bet there’s a new architecture to find that is as big as a gain as transformers were. I’d look for a mega-breakthrough and use the [current] models to help me”

On its own, this looks like a harmless prediction- technology has always had breakthroughs with previous learnings rocketing us into the future. However, this falls short of some of the claims of Sam Altman and others that we will achieve by end of the year or the next. But this doesn’t make sense. If LLM is the breakthrough, and we can achieve AGI, then why do we now need another technological breakthrough?

The Scaling Wall Nobody Wants to Talk About

For the past several years, the dominant religion of AI research has been scaling. The gospel was simple: more data, more compute, bigger models, and intelligence would emerge. This wasn’t entirely without merit. GPT-2 to GPT-3 to GPT-4 showed remarkable qualitative leaps, and the research community pointed to smooth, predictable scaling laws as proof that the path to AGI was just a matter of building bigger.

That narrative is quietly unraveling.

The gains from scaling are demonstrably flattening. OpenAI's own GPT-4 to GPT-4o improvements were largely about efficiency and multimodality, not raw reasoning leaps. The jump from GPT-4 to o1, and now to o3, required an entirely different paradigm - test-time compute, where the model is given more time to "think" through chain-of-thought reasoning before answering. That's a meaningful architectural shift, not a scaling increment. And crucially, it comes with enormous compute costs that scale poorly in production environments.

If brute-force scaling were sufficient, we wouldn't need that pivot.

Did Altmas finally admit to the limitations of current LLMs?

When Altman says he believes there’s “a new architecture to find as big as transformers,” he’s not making an optimistic prediction. He’s diagnosing a limitation.

The transformer architecture, for all its success, carries fundamental constraints. It processes context through attention mechanisms that scale quadratically with sequence length. Workable but not elegant for the kind of continuous, unbounded reasoning AGI would require. More critically, transformers are stateless between inference calls. They have no persistent memory, no ability to learn from new experiences without retraining, no model of the world that updates in real time. Every conversation starts from scratch.

What CEOs Say vs. What They Fund

Perhaps the most telling signal isn’t what AI leaders say in headlines, its where they quietly direct resources. Like I said above, the hiring of traditional developers by Anthropic, investing heavily on interpretability research are not the investments of an organization that believes it's twelve months from AGI.

The gap between the press release and the org chart is where the truth lives.

None of this means AI isn't consequential, transformative, or worth taking seriously. It clearly is. But there's a meaningful difference between a powerful tool that is reshaping industries and an artificial general intelligence that matches or exceeds human cognition across all domains. I think its time we all hit the brakes on the hype train and see AI for what it is.

If you found this interesting, I write about tech, AI, and the gap between hype and reality. Connect with me on LinkedIn/akshayramabhat.

Subscribe now

Why are we going to space to train models?

One of the draw backs of making AI videos of cats doing pull-ups in gyms is that it takes a lot of energy to make one. With dreams of AGI on every tech bro’s mind, the biggest hurdle, it looks like, is hardware and not software. Some are calling it an “infrastructure crisis”. As foundational models scale exponentially, the demand for high-density compute is colliding with the hard physical constraints of utility grids, land availability, and cooling water resources.

If we can’t keep up with demand on earth, we are forced to look elsewhere - space. In fact, we’re already in the Orbital Compute Age with the launch of the Starcloud -1 pathfinder mission in November 2025 which trained a Google Gemma model on an orbiting Nvidia H100 as a proof-of-concept that Orbital Compute is indeed possible opening a new era where the bulk of AI training will probably migrate to the abundant sunlight and the cold vacuum of space.

Mach33 posts research papers and news about frontier technology with AI and datacenters, especially about Orbital Compute and Orbital Real Estate (yes, thats what they call them). Some experts are predicting fully functional model training infrastructure deployed and run completely in space. I’m not going to blame you for thinking this idea looks a lot like science fiction, and I agree, it does. But lets dig deeper and see what the experts are actually saying.

The 3 main pillars for the transition

According to researchers, there are 3 critical pillars supporting this transition from terrestrial compute to orbital compute.

1. Orbital Real Estate: This market is segmented into specific regimes - Sun Synchronous Orbit (SSO) for near-term deployment due to its continuous solar exposure and High Earch Orbit (HEO) or Lagrange Point 1 (L1) for long-term, terawatt-scale training clusters.

2. The Energy Equation: Hardware costs are being rapidly offset by falling transport costs, driving the Orbital power costs down from the current $12 per wall for terrestrial systems to just $6 to $9 per Watt.

3. Thermodynamic Engineering: New research is overturning the historical misconception fo the “cooling constraint”, demonstrating that heat sinks in deep space offer superior thermal rejection capabilities for high-flux silicon (like the Nvidia H100 GPUs).

Lets explore each of these pillars in detail.

How much does it cost today?

The benchmark cost to deliver one Watt of power to the processor, excluding the compute hardware itself is, currently, $12 per Watt. The biggest cost drivers are -

• Utility Solar Plants: The generation component costs $1.0 - $1.6/W

• Grid and Substations: Interconnection costs range from $0.2 - $0.5/W

• Mechanical and Electrical: This is the bulk of the cost and it includes switchgears, UPS (and batteries), cooling loops and so on, costs $7 - $12/W.

The last point involving cooling infrastruture is truly one of the biggest motivations to move the training to space. The junction temperatures of GPUs can reach roughly 95°C which pushes air cooling to its physical limits and hence we conume up to 4 million liters of water EVERY DAY, enough to supply 300,000 PER Data Center.

This is obviously unsustainablt. The trajectory then becomes clear - to keep up with demand, we need a completely radical cooling system and space has the answer.

Where exactly can we put the datacenters?

The fundamental economic advantage of space is the solar multiplier. Space-based solar panels operate in a vacuum eliminating atmospheric absorption, scatrtering and weather related inermittency.

Lets get into more details on those pillars we talked about earlier.

The orbital realestate is essential split into 3 categories -

1. Dawn-Dusk Sun Synchonous Orbit (SSO) - This is the near-term recommendation or “starter orbit”. In an SSO, the satellite’s orbital plane rotates at the same rateas the earth orbits the sun (about 0.98° per day). By selecting the correct inclination and an orbit of about 600-800kms, we can achive perpertual sunlight without ever dipping into earth’s shadow.

   Moreover, since we’re technically in space, we get 8.2x cost-normalized solar index compared to earth which more or less eliminates the need for massive battery buffers.

   However, there’s a huge limitation with selecting this orbit. Since we’re looking to aim for near 24hr sunlight, the specific orbital corridor is very narrow - roughly 1% of the Low Earth Orbit’s (LEO) usable volume. As more operators rush to claim these high-value orbits, congestion becomes a critical concern. Collision risk and debris mitigation strategies will eventually cap the scalability of this regime. We need to go higher for long-term scalability.

2. High Earth Orbit (HEO) - HEO offers more than 500x the volume of LEO, providing almost infinite room for expansion with satellites experincing very long periods of sunlight, almost 95% in most cases. The problem with this orbit is that the environment here exposes electronics and solar panels to intense, high energy partlces due to the orbit’s location on the Van Allen Radiation Belts. This means we need more shielding of the components to avoid asset degredation.

3. Sun-Earth (L1) - This is the “endgame” location for orbital compute. Located 1.5 million kilometers from Earth, L1 is a gravitational saddle point where a satellite can maintain a fixed position relative to the Sun and Earth. This offers a continuous, unshadowed solar access and unconstrained physical volume. There’s no radiation belts that warrant the need for more shielding. The only biggest issue with computing this far away from earth would be the latency, which to be honest, is going to be a factor for all these orbits.

Alright, what about the cooling then?

On Earth, data centers cool chips by pumping water over heat sinks. In space, heat must be rejected via radiation. The power radiated is proportional to the fourth power of the temperature.

Space has a background temperature of around 3 Kelvin. Which means, instead of rejecting heat into ambient environment temperature of 303-313K, space’s effective sink temperature is VASTLY lower than Earth’s. Maintaining the radiator at a high termperature (60-80°C) while facing a near absolute zero background creates a massive heat flux potential that is unmatched in terrestrial systems without active refrigeration.

The challenge then turns from heat to physics. Radiators are large structures. The way we measure effeciency is with Areal Density (kg/m²) - how heavy the radiator is per unit of cooling area. For context, current honeycomb radiators weigh ~19 kg/m².

Next generation radiators are proving to be vastly more effecient. With deployable fabric radiators, flexible loop heat pipes, and oscillating heat pipes are demonstrating densities of less than 6 kg/m².

Space cooling is a closed-loop, water-free process. Heat is rejected as infrared light, leaving zero thermal pollution and consuming zero water.

So whats stopping us?

It all comes down to the price. Reaching HEO requires significantly more energy than LEO. However, Starship’s on-orbit refueling capability flattens this cost curve. We’re waiting for the price of rocket launches to come down, especially Starship. Currently, the launch on a Falcon 9 costs about $2000/kg. Starship is poised to bring it down even more at appox. $100/kg which is a HUGE improvement. Considering just a few years ago, we were paying upwards of $10,000/kg, the launch costs are coming down rapidly and is projected to continue to do so opening up the possibility of Space Data Centers very soon.

Orbital compute is not only a necessity but also a relief valve for Earth’s digital industrialization. Paving thousands of acres of land, consuming millions of liters of water and straining the power grids is not a long term solution. At the very least, these space data centres are kickstarting a new revolution in material science, something that works not just on earth but anywhere in space.

P.S: I researched the topic by reading Mach33’s articles on their website. Huge thanks to the Mach33 team for sharing their knowledge.

For the last few years, advancements in AI as we know it is not just a result of a series of revolutionary improvements around Large Language Model learning or the underlying development of neural network transformers but a series of almost magic-like innovations in the chip manufacturing techniques.

In 2019, as I was building my custom gaming PC, there was only one CPU manufacturer which was capable of cramming in insane amounts of permance in a 7nm architecture and it wasn’t Intel. AMD was one of the first customers of ASML’s new Extreme Ultra Violet (EUV) Lithograpy machine which could literally “copy” billions of transitors into a small chip. At that time, and even now, if you’re building a PC, you will ALWAYS see AMD as the top recommendation for your CPU chip. Since then, ASML has invested heavily on its new High Numerical Aperture EUV machines which has just started shipping in 2025. These machines can create 3-5nm chips which are just insanely powerful and effecient.

Now, why is progress in chip manufacturing so important for AI development? Well currently, training a model takes a ton of time and energy. The infrastructure needed to run such a model is almost impossible without a highly effecient memory system. It isn’t impossible to run a model on a device like your phone but the contraints are too high. A standard smart phone can run at most a 2B parameter model and thats not good enough for mainstream use. Thats why we use APIs in products like Keyword Scouter. If we could host a model on your browser and keep data completely local, we would.

Now the big question, what next?

To understand what’s next for AI, we need to understand what’s next for the chiops that are powering the AI.

We’re all aware of Moore’s law - “the observation that the number of transistors on a chip doubles roughly every two years”. Against all odds, it has been held true for decades but we’re now seeing signs of phyical limits. Like I mentioned earlier, the High NA EUV machines are just getting shipped. These have a NA rating or 0.55 instead of the earlier machines which had a rating of 0.33. ASML is actively working on a Hyper NA machines which are capable of NA greater than 0.75. This enables the machine to print even smaller features, sub 1nanometer, into the chip processes. These machines are expected to be shipped out in early 2030s and big FABs like TSMC and Sumsung will probably invest $20 to $30 billion on each plant to build these sub 1 nm chips. After this, we’re hitting the physical limit of whats possible with the current method we have building chips.

A silicon atom is only 0.2 nanometers thick. In the <1nm chip archtecture, we’re talking transistors that are just a few silicon atoms wide. There’s only so much deeper you can go without literally breaking an atom in half (and we don’t want that for obvious reasons). So where do we go from sub 1nm?

The smarter minds than us are already hard at work on this. Instead of positioning N-P parts of transitor side-by-side, scientists are already experimenting with great success the vertical arrangement of N and P type layers. Additionally, we could see ourselves move completely away from Silicon altogether and towards Graphene or Molybdenum which can be layer just 1 atom thick. These technologies might enable us to double or triple the number of transitors possible with a 1nm architecture from billions to literally trillions of transistors in a single chip unlocking some of the greatest computational power in our hands.

Full models on your phone or will there even be a phone?

A chip with trillions of transitors on a chip could mean you can host a full Gemini model on your phone. But why will we be still using a phone when you can cram everything into the rims of your glasses? AR/VR technology, especially, AR glasses will be able to cram the performance of a phone into the frame of your prescription glasses. With display technology rapidly advancing, it is a very likely possibility that the phone we have in our hand will be a thing of the past.

With these unimaginable processing powers, a lot of medical and real world simulations will be done and analyzed in seconds instead of days or weeks like we do now. With the data of the world in a chip possible, simulations will more and more resemble real world activiy, which scientists now call “World Simulators”.

Final thoughts

“Any sufficiently advanced technology is indistinguishable from magic.” - Arthur C. Clarke

If you feel like AI is magic, I don’t blame you, it probably is. With the improvement in technology forecasted in the next 10-20 years, it will likely be magic itself. Cheers to 2026 and here’s to another year of AI excitement 🍻

How are you preparing for the next semiconductor decade? I’m curious to know what you think.

Open AI has been almost bragging about the amount of investment they’ve been getting from, well, almost everyone. Microsoft, PwC, Thrive Capital, Nvidia and the list goes on and on. Recently, they even announced a $500 Billion (yes, with a B) Stargate coalition investment fund. No one in the history of tech companies got this amount of investment, this kind of support, and this level of attention. The US stock prices for all companies tied with Open AI also surged to the moon. Nvidia known for some of its advanced chips is the primary supplier to Open AI and other companies working on their own LLMs. Nvidia is also the biggest supplier of these chips to China. The US government has been limiting what chips Nvidia can sell to China so as to cripple China from advancing in their own AI frontiers. China winning the AI war would be a disaster to the US. And that’s not counting the loss of soft power over the world which the US currently flaunts with its tariff threats. China had to be slowed down even if it meant sanctioning them from buying the latest Nvidia chips.

Now imagine the shock the industry felt when a Chinese hedge fund named High-Flyer with AI research as a side project comes out of nowhere with one of the most powerful models in the world, beating the Open AI’s most advanced, O1 model in most benchmarks. According to some sources, it didn’t even cost $6 Million to build it. Deepseek’s reasoning model, the R1, employs some of the most groundbreaking technological breakthroughs to achieve this. The reasoning model employs Chain-of-Thought (COT) prompt engineering technique, Inference Time Compute scaling and many other state-of-the-art advancements to shift the perception of AI in almost every way. All this, for a fraction of the cost. Currently, Deepseek’s R1 model costs just 93% of the price of Open AIs latest O1 model. And that’s not even the best part.

Deepseek’s R1 model is Open Source. You can download the model and install it on almost anything and run it locally without it ever reaching out to the internet. There are instances of Deepseek running on Raspberry Pi. You can even host it on your phone if you want to. Imagine that, the world’s most powerful AI model is open source, spilling its secrets to everyone in the world. An embarrassment to Open AI, which is anything but Open. To make things worse, Deepseek’s app is currently No.1 in the app store. The stock market lost Billions over this news and rightfully so. This is not just a win for Deepseek. This is a win for open source; this is a win for anyone that thought you need billions of dollars to build something great.

Now, will all this context, with the understanding of why this is such a big deal and with the importance of the model being Open Source, we now know why Deepseek is such a big deal. Whoever hosts the model, makes the rules. So that’s exactly what Deepseek.com is. The website itself is hosted in China, abiding by China’s censorship policies and data mining rules. Of course it won’t talk about Tiananmen square, Taiwan, Tibet or any other geopolitical controversies involving China. Its not a surprise that a model hosted in China would be pro-china. Open AI is rightfully pro-american with its own set of sensorships and rules. Atleast with Deepseek, you can host the model on your own and viola no more Chinese censorship, or Chinese data tracking.

We need to see these LLMs for what they are - technological marvels. Judging them by their political beliefs is very primitive, almost childish. Deepseek’s open source model should be embraced and should be celebrated. Something our Americal friends can learn a thing or two instead of spending Billions on the next big thing.

Microsoft’s Willow moment?

A couple of weeks ago, we saw Google have a major breakthrough in Quantum Computing with its Willow chip. They’ve come with a new way to decrease the number of errors as they add more qubits to their calculation. This is huge because for decades adding more qubits always introduced more errors. The stage is now set for Quantum Computing to advance rapidly while reducing the errors and to be used in real-world applications as we’ve all been waiting for decades.

However, we’re not fully there yet. Although we do have to acknowledge this breakthrough, Google estimates atleast 5 -10 years before we see the impact in the real world.

There’s also another company working on Quantum Computing. Microsoft. And they’re close to having a big moment of their own. Apparently, they’ve taken a different approach to this and rely on something called Topological Qubits. According to Microsoft, expect a mojor announcement on this in 2025. Interesting times ahead.

Super Agents

We’re not there with Artificial General Intelligence (AGI) just yet. However, there are signs that we are close. Tech leaders like Sam Altman and Mark Zuckerberg have been teasing us about something they’ve been working on. Super Agents. These aren’t your regular ChatGPT chatbots, these are AI agents capable of learning human habits, take decisions on their own and just overall be a great help in almost all industries. For example, a Super Agent can let you know when you’re about to run out of detergents and order a box on its own without you ever asking for it. When Mark Zuckerberg recently said they are looking at replacing mid-level software engineers in his company, this is what he was talking about. AI Super Agents coding and deliver on their own. Scary times ahead for Software Engineers.

AlphaFold

AlphaFold from Google Deepmind has been making some news lately. The team has been working on AlphaFold 3 where they try to understand how drugs bind to proteins and other biomolecules. This is extremely useful in developing new patterns in early-stage detection of life-threatening disease. The Deepmind team has been working tirelessly to use AI in various real life applications like creating plastic eating enzymes. Over the past few weeks, Google Deepmind has been making their AlphaFold 3 model open source which has paved the way for rapid development. Expect more accuracy and more medical breakthrough from this in 2025.

AI is advancing really fast. Sometimes faster than most people can keep up. This is how ground breaking technology moves and I, personally am very excited to see whats in store in the near future. Follow more of my thoughts on x.com/akshay

When I began my career, this was India 2012, internet had finally reached the farthest corners of the town I was living in. I had just started working as a Quality Assurance tester for a small company in the town. The company developed products for Flea Markets in the US. With no prior coding education or experience, the best I could do was test a products that were being built. I’ve always had an itch to build products since I was a child so I felt the best move for me was to learn to code. Touch screen phones were becoming a thing and responsive designs were being talked about in the tech world. It became evident to me that we need to get the local legacy businesses online and quick.

The Learning Phase

I learned coding in evening private classroom sessions with 10 other people. I started with C# .Net since I wanted to build websites. I then changed my job role from QA to Developer at my job so I can get some real world experience. I learned quite a lot working on projects than I did just reading about C# or .Net. I also took up some front end technologies like javascript and jQuery.

My First Freelance Project

By 2014 I was able to build a website on my own. I started looking at business, specifically teaching institutes that didn’t have websites. This was the time when modern website builders like Squarespace or Webflow didn’t really exist. Websites were built the old fashioned way. Up until this point, I always thought coding was the hardest part. I realized quickly that finding a customer was even harder. I did not give up though. I make cold calls, I talked to friends and their friends, I created my own website to advertise. I even used facebook ads. But none of them worked. I then went to a local business myself and waited till the owner had a few minutes to talk. Then I did the hardest thing I had ever done in my life - talking to a complete stranger and trying to sell him something. He politely said he wasn’t interested and so I moved on. I tried a few more businesses and finally I got my first project. It was a computer training institute and the owner delegated his son (who was the same age as me) to work with me and coming up with the website design and requirements. The budget was *very* tight but I didn’t mind. I needed this project and did everything I could to make this work even if it meant I was losing money and time on it. I was paid half upfront and the rest later when it was done. Success.

A Few More Deals

I got a few of more projects after this. Not all of them were as smooth as the first one, though. There were times when the customer lost interest in the project and I walked away mid development. Sometimes the customer would get too pushy with unrealistic timelines. In the end, I learnt a lot more about business than I would have ever learnt reading books.

What I Learned

• The customer has no idea what they really want. Its up to you to learn about their business and figure out what they need. As I got more projects, I started spending more time with the customer learning about their business instead of getting straight into designing the website.

• The customer is not always right In the initial days I spent a lot of time just agreeing to what the customer thought they wanted from me by just being too scared of losing the business. It was important to set boundaries and expectations so we don’t disappoint each other in the long run.

• Value trust and accountability Word of mouth was the number one reason I got more projects. I had to make sure I went through with my promises as much as I possibly could. I learnt that it is VERY important to communicate with the customer as often as you can to keep them up to date with the progress. It was also important to let them know of any hickups in the projects ASAP instead of spending time trying to solve it.

After just over a year of freelancing, in 2014 I decided to move to a bigger city where I joined a more reputed company and worked on a lot more complex and challenging projects. I then went on to work for some of the biggest names in the tech industry like VMware working on some of the complex problems and managing some big projects. I then moved to Canada and currently work for Workday where I learnt a lot, understood the western market a lot more and continue to look for new and exciting opportunities.

Find me on X at @akshay_a_r

Lets connect!

If you're still doubting if AI is a bubble or not, you should checkout v0 by Vercel. This thing is still in its infant stages and you’ll be shocked at what it can do right now. If this is how the beginning of AI coding looks like, I can’t imagine how its peak will be.

I had an idea of creating an anonymous survey app.

Here's the exact prompt - "create an anonymous survey application. I need a rating system from 1 - 10 in terms of satisfaction. And then an anonymous comment box right under neath it. A question will be asked each time and then move on to the next question when answered. Come up with something pleasing to the eyes"

This is what it came up with -

And its not just an image. It works too! With validation and everything! Check it out -

This is possible thanks to V0’s amazing coding ability. Its not the best but it is better than you’d ever expect it to be. It came up with the logo, the header, footer and everything else all on its on. It also organizes and codes pretty well too. The code is readable and easy to port over to your favorite IDE if you want to. You can also ask it to modify anything you want or improve the app by just talking to it. Its amazing.

If you ever had an idea in your mind but just couldn’t put them into words, this tool might just be it for you to get started on your vision. Check out V0 by Vercel

Thanks for reading. Follow me on X for more at https://x.com/akshay_a_r

When my new accountant wanted some of my old tax receipts, all I had to do was hop on to my OneDrive, search for it and email the document straight to him. Pretty awesome. However, there’s a problem. I cannot store anything more than a few documents and some media without getting disk-space anxiety.

I kept adding documents and important photos and videos and I eventually ran out of space. I had a few options here-

1. Get a Hard Disk and put all my less priority files in there. There are a couple of problems with this. A hard disk is another Pen Drive but bigger AND I need to keep a track of the data on them. Also, how many hard disks will I keep buying? Where will I keep them? How will I take them all with me if I need something from there on the go? Perhaps, a better cost-effective alternative is to go with a cloud storage subscription?

2. Subscribe for a premium plan for a few bucks. It worked - for a while. I got 50 gigs of extra storage on my OneDrive account for a couple of bucks a month but with extra storage comes extra files. I quickly filled up the additional 50GB of storage, leaving me with limited space for my growing collection of documents, photos, and videos. I needed a more scalable and secure solution to store my important files. But, how can you trust your personal data you upload onto the cloud storage companies. They might be safe today but what about 5 years from now? What happens if your Google or Microsoft account is suspended? You have a LOT to lose. I needed something with more storage and more importantly something under my control.

3. NAS. Something the big boys at the enterprise companies have but they are now affordable enough to setup in your own home. Lets see why this was a game changer for me.

NAS, in case you didn’t know, stands for Network Attached Storage. Simply put, its a bunch of hard drives that lives in your closet. Like they say, Cloud is just someone else’s hard drive. The fancy people at Google and Microsoft have Data Centres. A closet works just fine for you and me.

For the upfront cost, the fun of setting it up and the customization it offers, it is well worth the investment. I’ll explain why.

A NAS from Synology comes with a Linux operating system of its own and its whole purpose is to serve you the data and no nonsense. You can get a really good WD 12 TB (yes, terabytes) hard drive on Newegg for like $200CAD. I have 2 for redundancy just in case one of them fails. The NAS itself is a Synology DS923+ for under $600CAD. There are cheaper options if you don’t want to future proof it with more drive bays. The storage drives with the Synology OS layer attaches to your network like any other network drive. This opens up a ton of possibilities for customization.

Media Server for Movies, TV Shows, Photos and everything else. Synology offers an application called Synology Photos which you just download on your phone and it automatically backs up everything you have to your NAS. You can also create your own collection of 4K movies you want on your NAS and play them using Plex. Plex is just Netflix but connects to your NAS instead of the Netflix Server. And its free.

With a bit of understanding of Docker containers, you can setup an end to end system to download any Movie or TV Show you want by just adding its name to a NAS client like Radarr. It automatically looks up the movie on your favourite P2P client and gets the best quality straight to your Plex streaming service in minutes. I cancelled ALL my media subscription thanks to this.

Tailscale for Remote Access from ANYWHERE. You can connect to your NAS from ANYWHERE using layered network solution from Tailscale. Yes, this means you have the NAS added as a network drive on your laptop wherever you go. Again, for free.

I’ve been using my NAS for a year now and I’m loving it. There’s so much more you can do and I’m still adding features and automation to it. If you’d like to know more on how I set the various E2E NAS flows up, subscribe to my Substack and comment below.

Thank you for reading.