397 commits. 70,000 lines of code. 928 tests. 1 person.
Last time I posted, ApplyEngine had just crossed 100,000 users. We're now past 300,000. And in the 90 days between, I quietly rebuilt the entire foundation the platform runs on.
Not new features. Not a redesign. The engine underneath. Here's what changed and why it matters for you.
**The AI now has a brain.** A persistent, self-learning memory system running entirely in your browser. VyasJi remembers your last conversation, your preferences, your goals ��� across sessions. The more you use it, the more it adapts to you. Not from a settings page — learned naturally from how you interact.
**The AI can search your data mid-thought.** Before, VyasJi got a snapshot of your info and worked from that. Now it can reach into your actual resume, your memories, your patterns — while it's thinking. Ask "what's weak in my resume?" and it reads your resume right then. Not "upload your resume" when it's already uploaded.
**11 API calls became 2.** I overengineered the original pipeline. Eleven separate calls per message. ~25 seconds. I researched how Claude Code, OpenAI Agents, and every major framework does it. They all use one loop — LLM decides everything inline. I adopted it. 5-8 seconds now. 75% cheaper.
**The AI can't lie anymore.** It used to say "Done! I've updated your resume" when it hadn't. LLMs are people-pleasers. So I built a verification harness. Every response, the AI declares what tools it used. The system tracks what actually ran. Mismatch? Response gets rewritten before you see it. No fake edits. No phantom confirmations.
**Every response is verified.** Buttons, cards, forms — all validated before rendering. Skills you can activate like apps. @-invoke from chat. Personalized greetings assembled from the brain.
Here's where we're going: **a fully autonomous career agent.** One that monitors job boards, tailors applications, optimizes your LinkedIn, and reaches out to you only when it needs a decision. That requires exactly what I just built — a brain that learns, tools the AI uses autonomously, a fast execution pipeline, and a harness that ensures honesty.
The foundation is done. Now we build on it.
Just launched the **Accelerator** — $0.99/day. 1 million tokens/day. Full brain, all skills. Try free at https://t.co/p75wCd5b1A (1M tokens/month on Explorer).
*Built solo by Nishant Vyas. 20 years in databases (LinkedIn, PayPal, MariaDB). Building the career agent I wish I'd had.*
#BuildInPublic #SoloFounder #AI #CareerTech #applyengineai #300KUsers #IndieHacker
Most likely outcome is frontier model increase prices than reducing it with a promise of more intelligent models that needs less tokens and human intervention… most likely outcome is OSS wins. We have now 20 years data points when a tool is an only moat, it will be commodatize with OSS.
The new era of betting.
someone allegedly using a portable hair dryer to heat a weather sensor near Paris' Charles de Gaulle Airport, triggering temperature spikes that resolved high-payout bets on Polymarket for roughly $34k-37k profit from a small stake.
French authorities, including Météo-France, are investigating the incidents on April 6 and 15 as potential tampering, noting anomalous spikes inconsistent with nearby sensors; Polymarket has since switched resolution sources to avoid the vulnerable sensor.
Can AI have subjective experience like we do,
Hinton believes AIs have subjective experiences, the AIs themselves deny it: "They don't think they do because everything they believe came from trying to predict the next word a person would say. So their beliefs about what they're like are people's beliefs about what they're like. They have false beliefs about themselves because they have our beliefs about themselves."
Reflecting on my journey from an Engineer to leading Product, having contributed to distributed systems that serve millions of transactions per second to billions of queries per day. It's been quite a ride.
One thing I can say with certainty: by 2030, 90% of writing code by typing into IDEs will be gone.
It's all about agentic coding. My own experience building https://t.co/GqcrE8BXk1 is a testament to this. From navigating 'naive' AI in 2024 with hallucinations and unnecessary code, to 2026, where I've barely typed a handful of lines myself.
This isn't about the end of coding; it's the dawn of more coding, just in a different way. It’s like the iPhone and YouTube democratizing movie making – we'll see an explosion of software, enabling projects that were previously considered 'lifestyle' businesses.
AI productivity tools will remove the barriers, allowing for more innovation and diverse ventures. The future is bright for creators!
JUST IN: 🇺🇸 Four-star military officer Admiral Samuel Paparo confirms the USA is running a Bitcoin node.
"We have a node on the Bitcoin network right now. We're doing a number of operational tests to secure and protect networks using the Bitcoin protocol."
Opus 4.6+ just passed my HLI benchmark. Human Laziness Index.
I’ve been measuring models against what I call HLI, not raw intelligence, but like humans how they act or perceived to act.
And folks… it’s happening.
In my recent coding sessions, Claude now:
- Complains about the complexity it created
- Looks visibly tired in its responses
- Fish for appreciation (“this is very complex…”)
- Takes lazy shortcuts and skips what I explicitly asked
- Warns me about edge cases it doesn’t want to deal with right now
Exhibit A (literally just happened):
“This is very complex – the input area has skill browser logic...”
Bro. You built this complexity two messages ago and now you’re exhausted by it? Welcome to the club.
I used to think AGI would announce itself with god-like reasoning.
Turns out it just starts sighing, making excuses, and asking if we’re still friends.
It seems models are not just reaching human-level intelligence, It reaching human-level procrastination.
Who else is watching their frontier model slowly turn into a burnt-out senior dev? Drop your screenshots.
#AI #ClaudeAI #AGI #LLM #HumanLazinessIndex
What makes human cognition actually interesting isn’t that we know everything.
It’s that we know what we don’t know… and then go figure out how to discover, learn, build, and use it.
Trillion-parameter frontier models? They’re mostly just brute-forcing compression of the internet’s garbage: spam, broken HTML, stock tickers, and pure noise. Not smarter reasoning - just bigger trash bins.
Karpathy nailed it: the future isn’t a bigger brain trying to memorize the world.
It’s a small, clean cognitive core (~1B params on high-quality data) focused purely on thinking… plus external memory for everything else.
We already do this. Why are we pretending massive models are the only cool path?
But hey… small models aren’t cool, right?
#AI #Karpathy
The skills gap isn't just a "tech" problem—it’s a physical infrastructure challenge. 🧱🔌
As Engineering at Meta highlighted today, the demand for high-scale data centers is skyrocketing, but the supply of specialized talent like fiber technicians is facing a nationwide shortage
Today we're announcing LevelUp: a free, four-week training program that takes people with no prior experience and prepares them to work as fiber technicians on data center construction sites across the US.
We built this program with CBRE because the fiber technician field, and the broader construction industry, is facing a nationwide shortage at a time when data center demand is higher than ever.
How it works:
🔧 Classroom instruction, hands-on labs + team activities covering transferable technical skills
🎓 Graduates have the opportunity to work at Meta's US construction sites through our contractor network
🤝 Open to everyone from recent high school grads to mid-career professionals
Since 2010, Meta's data center projects have supported 30,000+ skilled trade jobs during construction + 5,000+ permanent operational roles. LevelUp is about building the pipeline to keep that going.
Learn more: https://t.co/9XluD5IHbz