Narayan

Backend & Systems Engineer | Building pipeline infra in Java & Spring Boot | Top 13% ESA Kaggle | Currently building

[Narayan @realnarayan_]

Just shipped v2 of Too Many Issues,a platform to instantly find high-quality open-source issues.
Built a heavy backend to make it work:
> Spring Boot ingestion engine
> Polyglot persistence (Postgres + MongoDB)
> Redis locks for AI rate-limiting
Architecture diagrams below👇

[Narayan @realnarayan_]

The goal was to build a highly resilient pipeline that respects GitHub's API limits using a dynamic Hot/Warm/Cold polling algorithm.

Source Code: https://t.co/uxrQx2VbtS

Check out the live platform, the v3 roadmap, and the architectural breakdown here: https://t.co/yAV1sVf3nC

[Narayan @realnarayan_]

Just shipped v2 of Too Many Issues,a platform to instantly find high-quality open-source issues.
Built a heavy backend to make it work:
> Spring Boot ingestion engine
> Polyglot persistence (Postgres + MongoDB)
> Redis locks for AI rate-limiting
Architecture diagrams below👇

[Narayan @realnarayan_]

@adxtyahq Great Job Bro, make Sure it passes all he tests before its merged.

[Narayan @realnarayan_]

This was a good read, even though it was an old paper good for someone starting and learning,
but I have to say anything @karpathy suggests is a really good read.

[Narayan @realnarayan_]

@NotebookLM I have been asking for this since ages and finally it's here 🥹.

[Narayan @realnarayan_]

@curlysaarthak Wow, Nice work

[Narayan @realnarayan_]

@herooffjustice I think when the algo changes, @elonmusk was talking about i.e including grok in it, may be then the people who need will see it.

Also to mention I find reading about someone's intuition or mind better than any resource,it's just facts coming out from someone who went through it

[Narayan @realnarayan_]

@NotebookLM The only thing I would suggest at this point is just make the UI a bit better ,this product is anyways the most valuable out of all AI products that have come so far.

[Narayan @realnarayan_]

@ScholarshipfPhd Hi

[Narayan @realnarayan_]

@athleticKoder If "From Scratch" projects count😐:
https://t.co/lhyKp2PSdF

If a kaggle competition counts 🫠:
https://t.co/2cz6eH7akn

[Narayan @realnarayan_]

@herooffjustice Am I?

[Narayan @realnarayan_]

The test seems to be quite easy, almost solved all the questions,
But how would they hire with so many people appearing, while the test set is mediocre 🥲?

[Narayan @realnarayan_]

Hold on I forget that I have TCS NQT tomorrow 🥲 ,
will be back to building soon 🥀

[Narayan @realnarayan_]

Hold on I forget that I have TCS NQT tomorrow 🥲 ,
will be back to building soon 🥀

[Narayan @realnarayan_]

BERT with Numpy:
My "BERT from scratch" project is a disaster. 😅 (In the best way possible).
I've successfully built the BERT Model, all the layers, and the tokenizer. It works.
What's Missing are optimizations like vectorize in numpy, stable gradients etc.
Repo 👇

[Narayan @realnarayan_]

@fidexcode Check this Simple but works really good for personal projects :
https://t.co/2TIAnknMBE

[Narayan @realnarayan_]

I think this is one of the best VS Code Extensions you can have , it's really helpful especially for a guy like like me who tends to forget my own ideas by morning.
Extension Name: Todo Tree 🌳

[Narayan @realnarayan_]

Check the Repo now: https://t.co/lhyKp2PSdF

#DeepLearning #numpy

[Narayan @realnarayan_]

BERT with Numpy:
My "BERT from scratch" project is a disaster. 😅 (In the best way possible).
I've successfully built the BERT Model, all the layers, and the tokenizer. It works.
What's Missing are optimizations like vectorize in numpy, stable gradients etc.
Repo 👇

[Narayan @realnarayan_]

Completed the Bert Model design (from scratch with #numpy ), made the Transformer_Block which i almost forgot includes a Feed_forward_network in the paper, hence the delay.

Will be working on the loss_func now and then run the exps I was talking about yesterday.
#DeepLearning

[Narayan @realnarayan_]

@herooffjustice This exam thing always sucks, always breaks the momentum and then you have to rebuild the momentum again. 😐

[Narayan @realnarayan_]

@neetcode1 I mean who doesn't 👀

[Narayan @realnarayan_]

@herooffjustice Bro is that your dual monitor setup 😯?

[Narayan @realnarayan_]

Completed the Bert Model design (from scratch with #numpy ), made the Transformer_Block which i almost forgot includes a Feed_forward_network in the paper, hence the delay.

Will be working on the loss_func now and then run the exps I was talking about yesterday.
#DeepLearning

[Narayan @realnarayan_]

Okay wrapped single head attention and implemented a manager class for multi-head attention mechanism.

That's going to be it for today, too much to process on a single day 😮‍💨 ,
Will make the final model assembly tomorrow and run few tests and experiments.
#DeepLearning