anish

System Design By First Principles What is System Design? Netflix: Millions of streams. Zero downtime. Uber: Peak demand. Smooth rides. Zomato: Dinner rush. No breakdowns. That reliability isn’t luck. And it isn’t just “good code”. That’s system design. When your app is small: - Few users - One server - One database Everything feels easy. But as users grow: - Traffic spikes - Data explodes - Failures happen Costs increase That’s when systems break unless they’re designed to grow. At its core: System design is deciding how your system should behave before you start writing code. It answers questions like: - Where does the data live? - What happens during traffic spikes? - What breaks first—and how do we recover? - How do system components talk to each other? There’s an important difference most of us miss: Coding asks: “Does it work?” System design asks: “Will it still work at scale?” Every real system is built from the same basic pieces: - Clients (apps, browsers) - Servers (business logic) - Data stores (databases, caches) - Network (communication) - Constraints (cost, latency, scale) System design is about connecting these pieces so growth doesn’t break everything. There is no perfect system. Every decision is a trade-off: - Speed vs consistency - Cost vs performance - Simplicity vs scalability Netflix’s design isn’t “best”. It’s best for Netflix’s problems. That’s the mindset system design builds. What I’m doing next: I’m starting to learn system design properly from first principles. I’m not an expert. I’m learning this as a beginner. I’ll share all I learn. Let’s learn this together and see where it goes.

@theEquinoxDev Good one 😋

yet another topic - "temperature." so, when using any APIs while working with LLMs, there is a parameter called "temperature." what exactly is this and what happens when we decrease or increase it ? ofc climate doesn't change :) to understand that, we need to look at how an llm generates a response. an llm doesn't write an entire answer at once. it generates one token at a time, and at every step, there are multiple possible choices with different probabilities. for example, if i ask an llm: "write a function to check if a user is authenticated." the model always has multiple ways to approach it. it could generate a simple middleware function, use a helper function, add error handling, or structure the code differently or anything else, whatever is possible in relevance of the request. each of them can be valid outputs, but each choice has a different probability based on the patterns the model learned during training. the word temperature controls how the model chooses between all these different possibilities. with a lower temperature, the model is more likely to stick with the most probable choices. this usually gives more consistent and predictable outputs, which can be useful when you want good code generation. with a higher temperature, the model explores more possible choices, which can sometimes produce more unique solutions. but the important thing to understand is that temperature doesn't change the model's knowledge or coding ability. it doesn't make it smarter either... it only changes how the model selects between the possible outputs it already has while generating a response. this is also why the ideal temperature depends on what you are building. for eg, if you are building a code assistant, lower temperature generally makes more sense as you want consistent and reliable outputs. we don't want the model to randomize different approaches every time for the same problem. but if you are building something like a brainstorming assistant, a higher temperature can be useful because you might want the model to explore more different ideas instead of always choosing the safest option. so temperature is not a setting for making the model "better." it is more about deciding how much variation you want in the responses. a useful way to think about it is: lower temperature -> more consistency higher temperature -> more randomness that's it for today :) happy learning !!

@Keshavdotdev Good going bhai 🙌🏻

@Anishdotcom Yes, just trying to explore and integrate via project

Today's read - part 2

@neuroquark Yesss 😋

@Anishdotcom let's gooo

Today's Read

@Keshavdotdev Yep do it bhai Btw you are working with bullmq and redis right for your project

@Anishdotcom Thanks for sharing, gotta read it too

@theEquinoxDev Bhai aap bhi share Karo articles jaha se aap ai ka padhte ho

@Anishdotcom hell yeah !!

@Daaniyahkhan/from-cron-jobs-to-bullmq-scaling-background-tasks-in-node-js-3e23c2c8baef" target="_blank" rel="nofollow noopener">medium.com/@Daaniyahkhan/…

@AmanSheikhKhan i am doing it already, this was a fun article i found online, so i was sharing it

@Anishdotcom Just go and learn the redis from first principles.

@Keshavdotdev @samirande_ yeah its a good read

Today's read thanks @samirande_ for sharing this masterpice ❤️

@sharpeye_wnl Ooh nice bhai 🙌🏻

@Anishdotcom its going well, getting to work on a lot of new things. currently working on an eval pipeline and hence exploring stuff around this

some stuff i got done today: - went through llm as a judge and how you can use if for evals other than golden datasets - went through langfuse evaluation docs - tested first flight 007, i might finish it - started writing a deep dive blog on oops, hoping to not scrap it this time lol

@Harry_The_Nerd big w sirji congrats

A win as an engineer. The solution I proposed at work got deployed, and it's working extremely well! It's related to AWS and AI agents... Phew. I can relax a bit now.

@theEquinoxDev I am loving this from you bhai Acchi revision ho jati hai 😋

another topic for you to learn about today is "prompting". while going deeper into how we interact with llms, i came across few-shot prompting, and i thought the difference between normal prompting and few-shot prompting was something i should share with y'all. usually, when we give instructions to an llm without any examples, it is called "zero-shot prompting." for example: "convert this customer feedback into positive and negative points." the model understands the task from the instruction itself and generates an answer. but sometimes, instructions alone are not enough. the model might understand the task, but the output format might not be exactly what you expected. here enters another type of prompting - "few shot prompting. " instead of only explaining the task, you provide a few examples of what a good output looks like. for example: input: "the app is fast, but finding settings is confusing." output: positive: performance negative: usability now when you provide new feedback, the model has a reference for the pattern you want it to follow. the important thing here is that this is not training. the model is not learning this example permanently, and its parameters are not changing. the example is only included in that specific request and helps guide the response for that moment. so the difference is simple: zero-shot prompting: give the task -> model figures out the pattern few-shot prompting: give the task + examples -> model follows the demonstrated pattern few-shot prompting becomes especially useful when the task is specific, the output format matters, or when explaining the exact behaviour you want is harder than showing an example. hope this helped you understand this concept a little better. happy learning :)

@Harry_The_Nerd Bhai aapka job ka time kya hai

Good Morning from this nerd ❤️ Day 99 of my 100-Days-of-Fixing-Everything Challenge Coding- Leetcode LLD (lil bit) Preparing for a certification (claude) - Mock tests on udemy + Gen AI Revising Web Dev (properly) Databases in Depth Published the last article about databases (part 3) Character Design practice - 20 mins Writing - Worked on my writing project + Did some editing stuff for the publication (a lot) + wrote a poem & an article on medium Gym maxxing done Anti-brainrot going goodie! Explored some non-tech thingies Life-maxxing going goodieee

@sankitdev Shuru karna hard hai Ek baat adat lag gayi to easy hai 😋

I used to struggle to hit 10k steps. today I did it almost effortlessly. (not exactly) Here's the one change that made it happen.

@Keshavdotdev A is solid

Which UI looks better to you — A or B? 👀 I'd love to hear your feedback! Feel free to share any suggestions for improvement too. If neither feels quite right, let me know what you'd change. 🚀

@shivamxtech Unplanned trips are the best 😋

Post match, we went to a waterfall 🤩. unfortunately forgot to take any pictures but it was such a nice experience. Had lunch and now we're heading back home. (btw this was completely unplanned 😌)