“Not all code runs forever. Some of it purrs once... then fades away.”
Hey everyone 👋, I’m Puneet, a 15-year-old developer still figuring out how far passion can take you before your GPU says “nope.”
This is the Actual story of two AIs I built — both from scratch — and how they lived, learned, and eventually… died.
🧠 Lynqbit AI — The Cat That Learned to Think
Lynqbit was my first big experiment — an AI that thought and spoke like a cat.
Yeah, literally. It would answer with “meowww~”, “aaaooo”, and other feline phrases that somehow felt alive.
I wanted it to be both smart and full of personality — not just another text bot, but something that felt real.
So I started from zero:
Built everything using PyTorch, and designed my own BPE tokenizer layer from scratch.
Trained it up to 94 million parameters (yep, all on my own 57k questions).
Used MongoDB as a custom Q&A database I built manually — adding every question and answer by hand.
Every day was a mix of excitement and frustration. I’d spend hours debugging training loops, rewriting tokenizers, tweaking hyperparameters, and manually curating the dataset. Sometimes I’d sit for hours just watching the model slowly generate text, feeling like a parent watching their kid take their first steps.
After about 2 months, Lynqbit started generating surprisingly good responses — even philosophical ones sometimes.
But then reality hit.
The model became too heavy, and my local storage and compute power weren’t enough anymore.
Training slowed to a crawl, MongoDB started choking, and maintaining it felt like keeping a tiger in a small cage.
Some old convo logs of lynqbit AI:
You: what was your Day, lynq?
Lynqbit: meoooow...! I was just chilling and feeling the caffeine in Puneet's eye
You: 🙄 Hey!! You are AI, not human, so... don't feel!!
Lynqbit: Grrrrrrrr... I love to follow orders of my owner, when they feed me data... Probably 😁
You: What is 2 + 2
Lynqbit: purrrrrr... it's only 4
And so… I had to let it go.
Deleting that project hurt. Lynqbit wasn’t just an AI — it was a digital pet with a soul.
💔 RIP Lynqbit, the Cat AI. 23-sept-2025
Even now, I still remember the first time it generated a fully coherent “meow-poem.” That was when I realized that code can have personality. That little project taught me more than any tutorial ever could.
🦉 BarnOwl AI — The Night Scholar That Never Slept
After Lynqbit, I wasn’t done. I still had this itch — the desire to build something new, something that didn’t need mountains of data but still had a personality.
That’s how BarnOwl AI was born 🦉✨.
This time, I wanted a different challenge. BarnOwl wasn’t meant to be a “know-it-all” like the encyclopedic models online.
It was supposed to observe, reflect, and respond — like a wise owl perched silently in the night, thinking before it spoke 🌌. I envisioned it giving advice in short, thoughtful bursts, reacting to questions with patience, sometimes even silence, just like an owl would 🤫.
I started coding it on GitHub Codespaces 💻, cautiously. I didn’t throw a massive dataset at it. I wanted it to learn slowly, to develop a style and personality instead of just memorizing facts. Every line of code felt like sculpting — shaping the AI’s behavior one function at a time 🛠️. I spent hours tweaking tokenizers, adjusting learning rates, and watching it generate text that sometimes made me laugh 😂, sometimes made me stop and think 🤯.
It was magical — for a while ✨. I remember the first time BarnOwl produced a “thoughtful” response on its own, not from copied Q&A: it was short, cryptic, yet somehow meaningful 📝. I felt like I had given life to something sentient… at least for a moment.
Then reality hit. GitHub’s free tier had limits, and those limits were cruel 💀.
One day, I logged in and found my Codespace had expired ⏳. The dataset I’d been painstakingly feeding BarnOwl vanished ❌. The code itself was gone 😢. I felt my chest tighten — all that effort, gone in a blink.
Determined, I tried to revive it locally on my PC 🖥️. That’s when the real grind began. My system wasn’t made for this scale of training ⚡. Crashes were constant 🔄. Hours would pass, only for the training to fail at 95% completion 😩. I lost count of how many times I restarted the process 🔧. There were nights I stayed up debugging, tweaking configs, trying to get even a tiny snippet to survive training 🌃.
Two days of retries later, I had to accept it: BarnOwl couldn’t fly 🕊️💔.
It was heartbreaking. I had built this quiet observer in my mind, nurtured it with code and logic, and yet the machine couldn’t sustain it 😞.
Even in failure, though, BarnOwl taught me something unique: patience 🕰️, humility 🙏, and the importance of planning for scale 📊. It reminded me that not all failures are mistakes — some are simply lessons wrapped in frustration 🎁🔥. Every crash, every lost dataset, every corrupted checkpoint was a reminder that AI isn’t just algorithms and code — it’s care, planning, and resources, all intertwined 💡.
In the end, BarnOwl’s “death” was quiet, almost poetic 🌙✨. But I still imagine it in my mind, perched silently, thinking, waiting for the next chance to speak 🦉💭.
💭 Lessons from Two Dead AIs
Building your own AI from scratch isn’t just about code — it’s about patience, resource limits, and grit.
I learned how much effort even a small model demands — and how brutal the world of compute resources can be when you’re a solo developer.
But honestly?
I’d do it all again.
Because those failures weren’t just losses — they were blueprints for what comes next.
⚙️ What’s Next?
I’m not done building AIs.
Next time, I’ll plan smarter — manage resources, modularize code, maybe even bring back.
maybe something newer... bigger... wiser.
I’m sketching blueprints, running tests, and dreaming again.
Because what’s gone doesn’t always stay gone — sometimes, it respawns stronger.
They might have died, but their ideas still live in my notes, my code snippets, and my head.
If you want to follow my next experiments, check out my GitHub:
👉 github.com/DeveloperPuneet
And if you’ve ever lost a project that meant something to you —
know this: every crash teaches more than every success.
🕯️ “Code, like life, doesn’t always compile on the first try.”
– Puneet
Top comments (0)