nndbg
★ Featured · on PyPIA neural-network interpretability toolkit for PyTorch & HuggingFace models. Eight analysis planes to answer the real question: "where does this model encode that - and why did it say this?"
AI/ML Engineer · IIT Bhilai
I'm Darsh Nandu - a 2nd-year DSAI student who builds AI systems from scratch, pokes at model internals, and ships everything to GitHub. Currently deep in mechanistic interpretability.
Things I've shipped - libraries, agents, and systems. Built from scratch, because that's the fastest way to actually learn.
A neural-network interpretability toolkit for PyTorch & HuggingFace models. Eight analysis planes to answer the real question: "where does this model encode that - and why did it say this?"
The taste section. Three lanes, alternating directions - hover to pause and read the commentary.

Album
Views
Drake · 2016
6 God hours. The CN Tower cover lives rent-free.

Album
Scorpion
Drake · 2018
Side A for the bars, side B for the feelings.

Album
More Life
Drake · 2017
"A playlist," technically. A classic, actually.

Album
After Hours
The Weeknd · 2020
Blinding Lights on loop while debugging at 2am.

Album
Starboy
The Weeknd · 2016
Look what you've done - I'm a motherless coder now.

Album
Purpose
Justin Bieber · 2015
Yes, unironically. Sorry is a perfect pop song.

Album
Astroworld
Travis Scott · 2018
Wish we could leave and go to Astroworld.

Album
Utopia
Travis Scott · 2023
In my chrome heart era.

Place
Japan
Mt. Fuji · 富士山
Bucket list #1. Trains on time, code on point.

Place
Kyoto
Japan · 京都
A thousand temples and zero bad photos.

Place
Tokyo
Japan · 東京
Neon, ramen, and the future - all in one city.

Place
Gujarat
Rann of Kutch
White salt desert under a full moon. Home turf energy.

Place
Statue of Unity
Gujarat · 182m
Tallest statue on Earth. Gujarat doesn't do small.

Place
Mumbai
Gateway of India
The city that never sleeps - maximum city, maximum energy.

The GOAT
Cristiano Ronaldo
CR7 · #7
SIUUUUU. The standard for what relentless work looks like.

Moment
Euro 2016
Portugal · Captain
Injured, off the pitch, still coaching from the sideline.

Era
Real Madrid Years
2009 - 2018
4 Champions League titles in 5 seasons. Unreal numbers.

Machine
Starship
SpaceX · Fully reusable
The biggest rocket ever flown - and they catch it with chopsticks.
"I like building things from scratch - it's the fastest way to actually learn them."
I'm a second-year B.Tech student in Data Science & AI at IIT Bhilai. I got into AI by wanting to understand how language models actually work under the hood - not just call APIs, but really understand the internals. That curiosity has stuck.
These days I split my time between research (NLP, mechanistic interpretability) and building systems - agents, RAG pipelines, inference platforms. Outside of code I'm usually reading about how the brain works, or finding ways to apply those ideas to make AI systems smarter.
Currently
What I actually use - from model training to deployment. A few things I've happily retired.
Open to research collabs & internships
Agents, interpretability, or anything AI - my inbox is open.
avg. reply time < 24h