
AI Engineering @Google Spanner | Ex: Microsoft | BITS Pilani | Lessons from 8 years in building large scale distributed systems (Views are personal)
Bengaluru, India
Joined October 2015
- Tweets5K
- Following1.1K
- Followers4.4K
- Likes16.1K
Pinned Tweet
If you are aiming for Big Tech & can work diligently for 10-15 hrs/week for about 3 months, read this tweet.
In this worst market, a mentee was able to 3x his CTC (~ 1 Cr) in India, by following this simple advice.
CAUTION:
> This is NOT the only way but one of the ways I've seen working.
> If you have less than 21 days, this guide is NOT for you.
High Level Plan (with resources):
(I will deep dive on each step in later tweets)
> You need to gain ground on DSA, System Design, LLD & your resume. Most big tech focus on these only.
> Prepare your resume first. You'll not be able to apply anywhere/talk about yourself without it.
Be thorough with each project inside it.
> Keep on applying to jobs that align with even 30% of your expertise.
Don't reject yourself.
Yes, apply first! Once your resume is done. Apply.
Recruiters take time to reach out anyway.
> Don't apply to ALL dream companies in the first iteration above. Maybe apply to one.
> Choose 1 platform for Data Structure & Algorithms practice.
I recommend Leetcode because of popularity.
Do whatever sheet you want to do but practice questions daily.
Take notes on where you falter.
Don't just think about the solution.
Implement or you will fail.
(More on deep DSA prep in another tweet)
> If you are new to System Design, read donnemartin/system-design-primer on github.
Read this 2 times.
It will give you a basic understanding of things you don't know.
Cover the breadth of System Design before going deeper.
Get Alex Xu (@alexxubyte) books. Both volumes.
Read them cover to cover multiple times.
Watch Jordan has no life YouTube iff you have time.
Practice mocks under time.
(There is more nuance here, will cover in another tweet)
> LLD requires you to code + design. Make a list of all the design patterns, ask AI to summarize with an example & quiz you on this.
If you have time watch Concept && Coding YouTube channel.
Read the concurrency basics. (Use AI)
I'll just try to spend less time here as compared to other things.
> Invest about 60-70% time practicing DSA, 25% system design, 5% LLD.
This will vary depending upon who you are.
> DO MOCKS. DO MOCKS. DO MOCKS.
Don't skip this. You'll fail the interviews.
> Once you have covered the breadth of all the topics above, just start applying. Ask for referrals.
You'll never feel ready.
Apply anyway.
> Prepare a log of your interviews. Don't make the same mistake again.
> Use mocks to calm your wits & make mistakes. Have fun in the interview.
I'll write a detailed draft for each of the sections above.
Let me know your questions below, will answer in the upcoming tweets.
Most of you'll never read till this line.
Like I said, this guide isn't for most of you.
Just for a select few.
If you are one of them, I wish you luck.
See you on the other side. 🚀
"Premature optimisation is the root of all evil".
Reminds me of how the early internet was served exclusively from single beefy machines.
You probably don't need to optimise for 1M concurrent users today.
In the AI era where it is tempting to add "just 1 more" feature.
Resist this, ship.
This is a note to myself, btw.
154
Very valid argument, however I am afraid that most of the "Educational videos" are just entertainment and provides an illusion of learning.
Most people wouldn't want to put the carbs.
the value prop of education focused videos on youtube, or even insta, or even instructional or how to videos is lower than it has ever been & will likely go even lower.
why would you spend time watching a broadcast one to many video that is potentially optimized to be lengthy for ad revenue when you can simply ask an llm to create visuals, 3d models, & precise/compact/personalized text to go with it??
claude can even generate a full video for you if you want synchronous consumption.
i wish i had this when i was in school, imagine learning about history by watching a personal generated documentary etc. these kids have it too good.
2
189
Abhimanyu Shekhawat retweeted
I just built a *functioning* React app by describing what I wanted to GPT-3.
I'm still in awe.
Wow! I still remember 11 years back I tweeted about Intuit’s process to get free food when they came to our college.
Been good here.
3
282
ASD-STE100 is really a game changer.
Thanks Andrej for the nudge.
Simplicity is really the ultimate sophistication.
We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks:
Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better:
Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better:
Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better:
Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work!
In summary:
- As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding.
- Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.
I randomly remember an article by @shaneparrish where he tried to model the day in 15 min timeslots.
Having an 8 hr work day will give you 32 slots by this design.
@ProfCalNewport in his book Deep Work mentioned that most people can focus deeply for a maximum of 4 hrs/day
Now, I have this innate urge to try a million things, pursue parallel tracks & absorb as much as I can.
Even after knowing all the above statistics, I try to extend my focus beyond 4 hrs (maybe it is shallow work)
I am thinking of experimenting with ruthless prioritization now.
That would entail me saying NO to good things to prioritize the greater ones.
It is personally difficult for me, because at each decision point, you see plausible branches drifting away from you.
It can mean delayed personal or professional projects, lost hobbies & drifting friendships.
But trying to do them all, doesn't work either. It is not a computational problem. Some human experiences and feelings lie outside the bounds of rationality.
It is a tragedy to apply the logical toolkits to things that are not governed by it. Like feeling content.
There is a bookshelf that I see today, with a million books inside it.
In this lifetime, I won't be able to pick all of them.
Maybe someday I'll make peace with this.
2
203
I was held hostage today by my tabs.
Decided to WFH today because the tabs get messed up when I switch from a 3 Monitors to 2.
Anyone in the same boat?
3
274
Recently I read a tweet from Slava Akhmechet & I felt reading something straight from 2009.
The subject of the tweet was quite useful but it was not written (written and not packaged) to intently make it catchy.
It was written in a tone of a friend talking to you, more natural, like hooking into someones stream of thoughts.
Humans are very good at subtly catching the intent. Most ingenious salesman don't appear to be selling anything. They appear to be your friend proposing you a solution.
So even if you have to sell someone, something, you must adopt a knack for sounding more humane.
This is a breath of fresh air for people like me who don't do too well with sensationalism & are happy to do ordinary improvements over an extraordinary lengths of time.
Remarkable things are often not a result of a single blow.
Thanks Slava for indirectly prompting me to resume putting my own thoughts here, maybe in my own imperfect, humane way.
3
196
Bengaluru is really small when you are avoiding someone but really big when you want to meet someone. 😭😂
1
198
Will be conducting an interactive workshop on Distributed Systems for EuroPython 2026 in Kraków, Poland.
I'm extremely passionate about building planet scale systems & had been working on Distributed Databases for 8 years now.
Grateful for the opportunity to speak at one of the most esteemed Python conferences in the world & learn from the best innovators in the domain.
This brings me a step closer to my vision of democratising access to technology.
Excited for the event.
See you in Kraków, Poland during 12-19 July.
1
8
517
Link to the abstract of my workshop:
programme.europython.eu/euro…
3
135
Had a great time talking about 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁 𝗛𝗮𝗿𝗻𝗲𝘀𝘀 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 at Large Scale Production Engineering, Google.
I gave an example of building a Financial Agent to analyse my credit card spends using Gemma 4 family of Local LLMs.
Everything running under 15 GB RAM on my laptop!
The agent was performing quite well & in some cases, at par with the frontier models.
This was a clear testimony for the importance of Harness Engineering to build effecting Agents.
Linking the YouTube talk in the following Tweet. Will link the repo soon.
Nice little touch here @nikitabier
Love the animation for liking the Tweet mentioning SpaceX
1
5
297














