Hiring our first dedicated RAG + memory engineer at glymph.
we capture who said and what's said in a room, on-device. turning that into memory people can actually use is the hard part. it's where we could definitely do better right now.
2-4 yrs. you've built retrieval systems and made them accurate, not just wired up a vector db and prayed.
generalist energy. takes a vague problem and ships.
bonus if you've touched TTS β voice-first output is next.
Hyderabad, in person, with the team.
dm me and bring something you've built and broken.
Day 196 :
bramma transcribed english cleanly but kept hallucinating telugu β confidently writing down words nobody said. we assumed it was the messy telugu audio, the accents, the code-switching. wasn't any of that. the model was just too small to hold the language. sized it up and the ghosts went quiet.
Built bramma for 6.5 months. posted none of it. that's the mistake i'm fixing today.
idea -> working device in that time: a puck that sits in a room, listens with consent, and turns what's said into memory the org can actually use - who decided what, and why.
None of it solo. @chaturrved was the first to believe and build it with me.
From tomorrow, the whole journey in the open. the wins and the broken days. every one.
Everyone keeps saying AI will replace programmers.
Honestly, it sounds very smart⦠till you actually ship software and live with it.
Yes, AI writes a lot of code now. I use it daily. Most teams do. Thatβs not the debate. The real thing people miss is that building software is not about making something work once.
The real work starts later.
Maintaining it. Changing it. Fixing things when something breaks at the worst possible time. And suddenly youβre stuck with a codebase that technically works, but nobody really understands.
Thatβs when you realise writing code is just 20β30% of the journey. The rest is judgment. Knowing what to build, what not to build, and when to say βthis is a bad ideaβ even if it runs fine today.
AI is a powerful tool. No doubt.
But it doesnβt own consequences.
You still need engineers who understand the code and the business well enough to stop mistakes before they become expensive lessons.
Don't you think that we are overhyping speed and forgetting responsibility?
"Value your time. It is all you have. Itβs more important than your money. Itβs more important than your friends. It is more important than anything. Your time is all you have. Do not waste your time."
@naval
If I could send my 18 year old self a message, it would have three parts:
1. Prestige is often mistaken. Follow curiosity instead.
2. There's no way to avoid hard work. It's not sufficient, but it is necessary.
3. Don't take your parents for granted.
The more you learn, the easier it gets.
The hardest part of anything is the beginning, when everything feels unfamiliar and complex. But learning compounds.
Each lesson builds on the last, making what once seemed impossible feel effortless. Keep pushing through the discomfort.
And you'll realize you're capable of far more than you thought.
The easiest way to control the urge to give in to your bad habits is to imagine how youβll feel after doing it.
Like, imagine the emotions that come after procrastinating or overeating.
Pre-experiencing the consequences helps your brain resist the impulse.
Why I always start with the fundamentals (even when students want to skip ahead) π
I read this student feedback last week that made me smile:
"..π° ππππ ππππ πππ πΆππ πππ π, πππ ππππππππ ππ πππ ππππ ππ π°π«π¬, π π ππ'π 'πππππ πππππππππππ', πππππ πππππ, πππ. πΎππππ π οΏ½οΏ½πππ πππ ππππ ππππππππ ππ πππ πππ π°π«π¬ πππ πππ πππππππ/πππππ πππππ πππππππππ ππ ππ π ππππππ, π° πππ ππππππππ ππππ π ππππ ππππ ππ, πππ ππππππ π ππππππ ππππ ππππ πππππ πππππππ ππ ππ π«πππππ, ππ πππ ππ - ππππ ππ πππ ππππππππππ ππππ πππππππ π π¨π πππππ ππππ'π ππ πππ πππππππ πππ ππ πππ π΅πππ πππππππ ππππ ππππππππ ππ οΏ½οΏ½οΏ½οΏ½ππππ ππππππ."
Yes, this is exactly why I designed the bootcamp to cover the fundamentals first.
Most people want to jump straight to Kubernetes, CI/CD pipelines, the "cool stuff."
I get it. But here's the thing:
If you don't know what problem a tool solves, you're just memorizing commands.
When this student - coming from Ops - manually set up a dev environment, they felt the actual pain.
Now containers will make sense.
Same goes the other way.
Developers who skip Linux and OS fundamentals?
They struggle later. They don't understand why their app crashes, why permissions matter, how processes actually work.
Both sides need the foundation.
Ops people need to understand dev workflows.
Dev people need to understand the system their code runs on.
I could start the bootcamp with "Here's Docker, run these commands."
But then it's just another tool to remember. Instead, we do it the hard way first.
We feel the friction. THEN the solution clicks.
(And yes, AI makes this worse. Now people copy-paste solutions without understanding anything. Then wonder why nothing works when they need to debug.)
Fundamentals aren't boring. They're context.
They're the difference between knowing HOW to use a tool and understanding WHY it exists.
Skip them, and you'll always be guessing.
π¬ What's a fundamental you wish you'd learned earlier?
An open secret about exercise is that initially everyone hates it, but after a while it becomes such a joy that you start missing it on the days you donβt work out.
Remember to just push through the initial phase and build it in a habit. Your future self will thank you for that.
i need to learn how to talk better when its technical bro
i got so much stuff in my head and then i say the most basic shit
maybe its sleep
never skip sleep
Donald Trumpβs 100,000 H-1B fee will choke U.S. innovation, and turbocharge Indiaβs. By slamming the door on global talent, America pushes the next wave of labs, patents, innovation and startups to Bangalore and Hyderabad, Pune and Gurgaon . Indiaβs finest Doctors, engineers, scientists, innovators have an opportunity to contribute to Indiaβs growth & progress towards #ViksitBharat. Americaβs loss will be Indiaβs gain.