Short Introduction

Manpreet Singh

AI/ML Engineer

I'm an AI enthusiast who blends hands-on industry experience with deep technical expertise.
After three years of crafting AI solutions in the real world, I pursued my passion further through a Master's in Computer Science, diving deep into machine learning and NLP.  At HyperMatrix Labs, I explored the fascinating world of Large Language Models, focusing on understanding their internal mechanisms and underlying behaviors.
I love tackling complex challenges at the intersection of research and practical applications, always looking for ways to push the boundaries of what AI can achieve.

Journey.

  • Education
  • Experience

Masters of Computer Science

Dalhousie University, Canada

2023 - 2024

Master's in Computer Science strengthened my understanding of core concepts in Machine Learning, Deep Learning and NLP, providing me with a solid foundation in artificial intelligence technologies.

Bachelors of Computer Applications

Guru Nanak Dev University, India

2016 - 2019

Started my CS journey here, learning core topics such as programming languages, software development, algorithms, data structures, computer architecture and operating systems.

Projects.

❮
Interpreting effects of Quantization on LLMs
This research project investigates how quantization affects the behavior and internal representations of LLMs. Our analysis focuses on three key areas: (1) model confidence and calibration metrics, (2) architectural redundancy and its relationship to quantization and (3) the interpretation of individual neuron behaviors under different quantization schemes.
Manifold Learning in NLP Embeddings
This project applies manifold learning techniques, including Isomap, Locally Linear Embedding, and T-SNE, to explore biases and learning patterns in NLP embeddings. By reducing the high-dimensional space of these embeddings, we visualize and examine the patterns and biases, assessing how effectively different manifold learning methods preserve crucial information.
❯

Get in touch.