Your Name

Adarsh

LLM & Vision-Language Researcher | Machine Learning Research Intern @ I2CS Research Group-IIIT Kottayam
Aspiring PhD in AI & Healthcare | Open to Research Opportunities

About Me

currently working as a Machine Learning Research Intern at the I2CS Research Group, IIIT Kottayam. My focus lies in applying deep learning to complex challenges in healthcare.

My approach is grounded in a first-principles understanding of AI. I work on deconstructing and implementing foundational models like Large Language Models (LLMs) and computer vision systems from scratch to bridge the gap between mathematical theory and practical application. This hands-on methodology is central to my work in developing robust and interpretable models.

I am currently preparing for a PhD in AI/ML and am actively seeking research opportunities where I can apply this foundational approach to solve high-impact problems. Feel free to browse my projects and articles to learn more.

Projects

Built-from-Scratch Language Model to Mimic Communication Style via Personal Chat Logs

An end-to-end project to build and fine-tune a custom language model capable of mimicking a unique communication style from personal chat logs. The process involved data extraction and cleaning, building a BPE tokenizer, training a decoder-only transformer from scratch, and implementing QLoRA for parameter-efficient fine-tuning. (Currently Developing)

Python PyTorch Fine-Tuning LoRA QLoRA
View GitHub

Reimplementing a LLaMA-style Transformer from Scratch in PyTorch

An educational implementation of a modern, Llama-like Large Language Model built from first principles in PyTorch. This project deconstructs the core transformer architecture, including hands-on implementations of Multi-Head Attention, Grouped-Query Attention (GQA), and Rotary Positional Embeddings (RoPE).

Python PyTorch LLM Deep Learning Transformers
View GitHub

Predicting Student Success with Machine Learning

A comprehensive data science project to predict student academic outcomes using a UCI dataset. This work involved the full pipeline: data cleaning, extensive exploratory data analysis (EDA) to uncover key trends, and feature engineering. Multiple regression models were trained and evaluated, with Random Forest selected as the best-fit model after addressing overfitting. A key part of this project was navigating the ethical challenge of dropping a statistically significant but potentially biased feature (gender) from the final model.

Python Scikit-learn Data Science Machine Learning EDA Feature Engineering
View GitHub

Publications

An End-to-End Sign Language Translation Pipeline from Static Gestures to English Using T5

N A Adarsh Pritam, Asha Kurian

Proceedings of the International Conference on Emerging Technologies in Computing and Communication (ETCC 2025)

The paper has been accepted and presented at an IEEE conference (2025). The paper will be published on IEEE Xplore in the coming months.

Blogs

Student Blog

A Beginner's Guide to Multi-Head Self-Attention in LLMs

August 26, 2025

LLM Series Transformers Linear Algebra
How do models like Llama and GPT understand context so effectively? The answer lies in multi-head self-attention. In this post, I provide a step-by-step breakdown of this foundational technology, translating complex theory into intuitive concepts.
Read on Medium →
BPE Blog

Build a Byte-Pair Encoding (BPE) Tokenizer from Scratch in Python

July 20, 2025

LLM Series
This is part of my journey to understand large language models from first principles. I’m building a LLaMA-like model from scratch, documenting each component in this blog series.
Read on Medium →
Student Blog

Predicting Student Success: Data Science and Machine Learning Insights

August 13, 2024

End-to-End Project Applied Data Science
A detailed article walking through a real-world data science project: from EDA and feature engineering to model selection and ethical considerations.
Read on Medium →
See More Blogs →

Open Source Contributions

For a complete overview of my open source contributions, check this GitHub search query. Below is the same information organized by year.

2025

    • Resolved UnicodeDecodeError in CSV reader by specifying encoding for cross-platform compatibility (FreeCodeCamp Tutorial)
    • Fixed typo in function docstring to improve clarity (FreeCodeCamp Tutorial)

Get In Touch

I'm actively exploring research opportunities and always open to meaningful collaborations in AI, machine learning, and LLMs. Whether you're a researcher, mentor, or fellow enthusiast, feel free to reach out!