MENU CLOSE
lukshya
supyal

MEET LUKSHYA↓

AI / ML Engineer · Open-source contributor

I build machine learning models and Python tools, and fix real bugs in the open-source libraries I use every day.

Let's connect
Luck can be freaky
Scroll to explore

Live GitHub activity

Building in public.

View GitHub profile
Illustration of Lukshya with sunglassesRemember, every model is a human opinion embedded in mathematics.
ABOUT ME

Hello! I'm Lukshya Supyal.

EVERY FRAME IS ART.

“Imagination is everything. It is the preview of life’s coming attractions.” — Albert Einstein

I am drawn to the space between imagination and reality—the place where an idea becomes something that actually works. I spend my time exploring LLMs, Transformers, and the mathematics behind intelligent systems, trying to understand not just what they can do, but why they work the way they do. I enjoy turning complex ideas into things people can actually use, experimenting with models, breaking systems apart, and rebuilding them in ways that make sense. I like to bend space and time—not literally, but through ideas, mathematics, and code.

Remember, every model is a human opinion embedded in mathematics.
Lukshya's resume
my resume, click me ↗
Retro computer sticker saying do epic shit, loading
still loading... :)

ACTIVE CONSTELLATIONTools that move ideas.

QUESTIONPROTOTYPEINTELLIGENCESHIP
LUKSHYA SUPYAL / AI · CODE · DATA · AUTOMATION
03 / CAPABILITY MAP

Skills in Orbit

Languages, frameworks and tools do not sit in separate boxes here. They revolve around the same job: turning an unusual idea into something useful.

Python
C++
Java
JavaScript
Python
C++
Java
JavaScript
PyTorch
TensorFlow
Scikit-Learn
NumPy
Pandas
PyTorch
TensorFlow
Scikit-Learn
NumPy
Pandas
Matplotlib
React.js
SQL
HTML/CSS
Jupyter
Matplotlib
React.js
SQL
HTML/CSS
Jupyter
Git
GitHub
VS Code
Linux
Google Colab
Git
GitHub
VS Code
Linux
Google Colab

07 / Certificate catalog

Certificates

Title -
Catalog
Keys -
04
Years -
2024–26
Press -
any key

From a student’s desk, I build software that learns
with curiosity & rigour, and a ship-then-sharpen approach.

Lukshya Supyal

AI/ML engineer &
Full-stack builder

Building since 2023

Studying
@BTech CSE (AI/ML)

Info

I’m a computer science student specialising in AI and machine learning. I like projects that start as a small itch, a video I want offline or a process eating my RAM, and end as tools other people can use. Each one below was designed, built and shipped by me, from the model to the interface.

Right now I’m building retrieval-grounded agents and chatbots that say “I don’t know” instead of inventing answers, and contributing fixes upstream to the libraries I use every day.

Timeline

  1. Chapter I

    2023 – 2024

    First tools

    Small Python scripts that scratched my own itches: a video downloader, then a model that watches my OS.

    PythonFlaskScikit-learn
  2. Chapter II

    2025

    Shipping for others

    BlueLink and the support agent: real users, schedules, retrieval and deploys.

    TypeScriptFastAPIRAG
  3. Chapter III

    2025 – Now

    Grounded LLMs

    Lua and Nova: chatbots that cite live data instead of guessing.

    LLMsAPIs

Selected work

YouTube Video Downloader

MediaPython
YouTube Video Downloader

AI Process Analyzer

OSMachine Learning
AI Process Analyzer

BlueLink

NewsWEB
BlueLink

AI Support Agent

AgentRAG
AI Support Agent

Lua: Anime Chatbot

ChatLLM
Lua: Anime Chatbot

Nova: Novel Bot

BooksLLM
Nova: Novel Bot

In other people’s codebases, I fix what’s quietly broken
with small diffs & full test runs, and a reviewer-first approach.

Open Source

Contributor &
Bug hunter

27 merged pull requests

Across
@18 repositories

Info

Open source is where I learned to read code written by strangers. I look for real bugs in libraries millions of people depend on, reproduce them, prove the fix with the project’s own test suite and keep the diff small enough to review over coffee.

These are six of the repositories I’ve landed changes in, from scikit-learn and Scrapy to pdf.js, Black and Hugging Face Diffusers.

Timeline

  1. Chapter I

    2026

    Reading strangers’ code

    Docs and examples first, in scikit-learn and pdf.js, learning how maintainers review.

    DocsPythonJavaScript
  2. Chapter II

    2026

    Real bugs

    Scrapy, Black and supervision: reproduce it, prove the fix with their tests, keep the diff tiny.

    TestingPython
  3. Chapter III

    2026 – Now

    Inside ML libraries

    Diffusers and Sentence Transformers: fixes inside the models people ship.

    PyTorchHugging Face

Merged upstream

scikit-learn/scikit-learn

Python★67.4K
scikit-learn/scikit-learn

scrapy/scrapy

Python★64.5K
scrapy/scrapy

mozilla/pdf.js

Javascript★53.9K
mozilla/pdf.js

roboflow/supervision

Python★51K
roboflow/supervision

psf/black

Python★41.9K
psf/black

huggingface/diffusers

Python★34.6K
huggingface/diffusers

Between experiments, I write down what broke and why
with honesty & detail, and a learn-in-public approach.

Field Notes

Writing &
Thinking out loud

Six notes in 2026

Published
@Blue Link

Info

Field Notes are the long-form side of my work: what I measured, what surprised me and what I got wrong. They’re written for the version of me from six months ago, which usually makes them useful to someone else too.

Mostly systems and machine learning: profilers, dataloaders, attention, Kaggle and the habits that make me a better engineer.

Timeline

  1. Chapter I

    Apr – May 2026

    Habits

    Reading papers above my level, and a semester without an IDE.

    ReadingTerminal
  2. Chapter II

    Jun 2026

    Measuring

    A Kaggle bronze and what attention really costs on a 4GB card.

    KaggleTransformers
  3. Chapter III

    Jul 2026

    Systems

    Three weeks inside a dataloader, and the GPU that was never the bottleneck.

    ProfilingPyTorch

Latest notes

The GPU Was Never the Bottleneck

Systems9 min read
The GPU Was Never the Bottleneck

Notes From Three Weeks Inside a Dataloader

Machine Learning11 min read
Notes From Three Weeks Inside a Dataloader

Attention Is Cheaper Than You Think (Until It Isn’t)

Machine Learning12 min read
Attention Is Cheaper Than You Think (Until It Isn’t)

What a Kaggle Bronze Actually Taught Me

Notebook7 min read
What a Kaggle Bronze Actually Taught Me

The Semester I Stopped Using an IDE

Notebook8 min read
The Semester I Stopped Using an IDE

On Reading Papers You Don’t Understand Yet

Notebook6 min read
On Reading Papers You Don’t Understand Yet

Have an opportunity or just want to say hello?

Let's make something
worth remembering.