Profile avatar
aayushdev.xyz
Applied AI | Deep Learning Student https://aayushdev.xyz
153 posts 7 followers 185 following
Regular Contributor
Active Commenter

Another new hire this week Things he's done so far: - vibe coded a product that had a api call - which had a backup call which had a fallback hardcoded backup call which hand another hardcoded fallback - It was a message to clean up - doesn't use org's git repos

I've definately got the dates wrong cuz I started in 28 oct

Day 31: - built an ai agent from scratch without using any framework, just the GenAI SDK with chroma db - didn't rely much on ai to write code for me, as i mentioned for the past few days - so things work, I know how, and the footprint is much smaller, just one module at a time - made as poc

Day 30 begins today: It feels like I haven't done almost anything in the past month

Day 29: - did some job applications - hit the gym - I restarted 3 days ago - did some work on new project we started this week - less vibe code means more mental presence and actually developing an understanding of things

Day 28: - reduced my vibe coding rate today - didn't feel so high achieving today, mostly worked on resume - making it into one page was harder, single page looks unskilled and tasteless

Day 27: - got a skill check from all the AI bots; most said I have a generalist profile and need to deepen my knowledge focusing on a specific field/skill - i am suitable for ai researcher/engineer role, data scientist, applied AI engg - my research profile isn't strong, doing MS with my

Day 26: - fixed my blog's deployment - found a cool blog: supaiku.com/attention-is... ; So I updated mine with a similar Hugo Theme: nostylesplease - fixed the project that keeps dying - it decided to breathe on its own so saved some efforts there

Day 25: - updated my Hugo blog - did some clauding

If you had one month to become a researcher in ML/LLMs What would you do? (this question is very subjective and vague, but I need some advice) #mlsky

Day 24: Got diarrhea; I was ill last Sunday as well but i did some testing with our major project, slept the afternoon and woke up to diarrhea 😑 I had Sabudana with dahi(sago cooked in chilli and some ground nut powder and curd (sweetend)) It's a common dish often cooked when fasting

Day 23 (late post): - went college - did a clean up of a service that kept breaking with Claude - went for a swim: it was rewarding; I can almost swim - 3/4th the breadth covered; more practice and i could do it

Day 22: Went to college today It is both a physical and mental toll Unlike their international counterparts they require us to write/print lab files and codes and make you run here and there 10 times before they agree to sign your files Anyways read up on some RPCs, RMIs and genetic algos

I wonder what happens to your obsidian notes once they enter your vault's void. youtu.be/pQy193-CFmc

🔊 Meet Orpheus: A breakthrough open-source TTS model that matches human-level speech with empathy & emotion. - Available in 4 sizes (150M-3B parameters) - delivers ultra-fast streaming - zero-shot voice cloning. - Apache 2.0 license Check it out: huggingface.co/collections/...

Day 21: What would be the fastest way to rewrite something as a microservice? This project has been a pain, it fails for all the reasons that aren't caused by me it was supposed to be a one month thing now it's been failing for once or twice every month. I'm in pain 😔 I need help

Day 20: Did some tiny fixes on our major project's API Some work at college Also yeah almost forgot, good news I got All india Rank 3876 in GATE DA

Day 19: Setup CI/CD thingy for my websites Setup blog using Hugo with breablog theme Afternoon naps become mandatory in summer -- specially when you gotta do by just a Fan Brain goes crazy when it's hot, rather than letting the neurons deplete into social media and reels ... I take a nap

I've been holding off applying to anthropic for so long now. I really wanna get into their interpretability team, my profile isn't strong enough yet.

🚀 Just dropped: SmolDocling-256M brings powerful OCR to your local machine! Only 256M in size but packs full document processing power.

This is a very tidy little RL paper for reasoning. Their GRPO changes: 1 Two different clip hyperparams, so positive clipping can uplift more unexpected tokens 2 Dynamic sampling -- remove samples w flat reward in batch 3 Per token loss 4 Managing too long generations in loss dapo-sia.github.io

Day 18: Late post. Was asleep most of yesterday. Diagnosed the major project we are finishing up on. Client didn't like the demo Many things are barely working, we need to improve them.