One year ago, around April 2025, I got my first real development client on Upwork.
At that time, I was still figuring out how to position myself, how to talk to clients, how to handle real production issues, and honestly, how to survive as a freelancer.
Since then, I’ve worked with multiple international clients across AI SaaS, education platforms, legal-tech, real estate, booking systems, dashboards, automation tools, and business websites.
A lot of my work was not just “build a page” or “fix a button.”
I worked on things like: fixing and improving Lovable-built apps writing backend systems from scratch working with Supabase, PostgreSQL, RLS, Auth, and Edge Functions building admin dashboards and client portals improving UI/UX for live products integrating AI personas and AI workflows setting up Stripe, webhooks, email flows, and third-party APIs debugging production issues that were blocking real users improving security, database logic, and deployment workflows helping founders turn AI-generated code into real working products
One thing I learned this year is that freelancing is not only about coding.
It teaches you communication, ownership, patience, speed, problem-solving, and how to stay calm when something breaks in production.
There were projects where I had to jump into an existing messy codebase and understand everything quickly. There were times where Lovable, Codex, or Claude generated something, but it still needed proper engineering judgment to make it reliable.
There were also moments where I had no perfect answer at first, but I figured it out by staying with the problem.
This one year gave me more real-world learning than any course could.
I’m still improving, still learning, and still building. But I’m proud that I’ve been able to work with real clients, solve real problems, and contribute to products that people are actually trying to launch and scale.
Going forward, I want to keep focusing on what I enjoy most: AI full-stack development, Supabase-backed apps, automation systems, dashboards, SaaS MVPs, and helping founders build faster with AI without losing technical quality.
Grateful for every client who trusted me in this first year.
And excited for what the next year brings.
2004 was a good year, but your Gmail address doesn't need to be stuck in it.
To say goodbye to [email protected] or [email protected] (or whatever you were into at the time), go to your Google Account settings and choose any name available. You'll keep your old username and you can sign in with both.
TurboQuant in plain English:
Think of a smart AI chatbot like ChatGPT or Gemini. When you chat with it for a while or give it a long document to read, it has to remember everything you said earlier. It stores that memory in a special notebook inside the computer called the key-value cache.
That notebook gets huge really fast. It eats up tons of memory (RAM) and slows everything down. On a phone, laptop, or even a normal computer, this means:
• The AI can only handle short chats before it chokes.
• It needs expensive powerful hardware.
• Responses get slow.
Google Research just released TurboQuant, a new compression trick that:
• Shrinks that memory notebook by at least 6× (sometimes way more).
• Makes the AI up to 8× faster.
• Does it with zero loss in accuracy (the AI is just as smart as before).
It’s like taking a giant photo file, compressing it to 1/6th the size with a perfect zip tool, and the picture still looks identical when you open it. No blurry edges, no missing details.
What this actually means for regular people:
• AI chatbots can now handle much longer conversations without slowing down or running out of memory.
• It works better on phones, laptops, and cheaper computers—no need for giant data-center GPUs.
• Future AI (including Google’s own models) will feel snappier and cheaper to run.
• The little animation in the tweet shows colorful bars (representing AI memory) getting neatly packed into a tiny grid. That’s exactly what TurboQuant is doing behind the scenes.
Bottom line: Google figured out a smarter way to make AI’s memory tiny and lightning-fast without sacrificing quality. This is the kind of behind-the-scenes breakthrough that will make AI feel way more practical in everyday apps soon. No magic, just really clever math that finally works perfectly.
First came the (1) Freelancer-for-hire era
Then came the (2) No-code builder era
We are now entering the (3) Founder-who-ships era
Few are building in this window. Most will wish they had.