Hi, I'm Edwin Chacko.

I'm a machine learning researcher and embedded systems enthusiast with experience in ML and systems programming currently exploring the intersection of AI and hardware!

Experience Timeline

April 2024 - Present

Machine Learning Researcher

McMaster University - ChemAI Lab

Led a cross-disciplinary team of chemists and engineers to develop a deep learning model for molecule elucidation, consulting Dr. Kylie Luksa for domain-specific knowledge in spectroscopy. Built and optimized a custom GPU rig, enabling the training of the model on 15,000 molecular samples over 20 hours, significantly reducing computational bottlenecks. Designed a multimodal architecture combining a fine-tuned CNN with an MLP for IR images alongside an LLM2Vec-based NMR text encoder, achieving 93% accuracy in molecule prediction and a 91% F1 score for functional group detection. Implemented SMOTE, data normalization, and augmentation techniques to reduce class imbalance by 30%, improving model generalization. Set the foundation for future deployment as a real-time tool for chemical analysis, with potential applications in drug discovery and material science

May 2023 - August 2023

Calibration Engineer Intern

VACS Calibrations Ltd.

Calibrated electronic and mechanical equipment, following the IEE and ISO17025 standards. Performed statistical analysis including standard deviation, uncertainty propagation, and regression methods to validate calibration accuracy and reliability, improving accuracy by 20%. Improved process efficiency by 10% and measurement precision by 15% utilizing insights developed from calibration data.

Graduation Expected May 2027

BASc. in Engineering Science, Machine Intelligence Option

University of Toronto

Relevant Coursework: Machine Learning, Data Structures and Algorithms, Natural Language Computing, Computational Linguistics

Projects

Spectro

TensorFlow, PyTorch, NumPy, Linux, Docker

Worked extensively with TensorFlow, PyTorch, and NumPy to create the Spectro workflow. Consists of 2 models, Spectro, an RNN + LSTM decoder, and j-IR-vis, a CNN + MLP.

Demo and Github Coming Soon!

Ultimate Chess

C++, CUDA, Linux, Docker, Postman, REST API

Hardware optimized C++ chess engine designed for performance. Utilizing techniques such as Bitboards, Piece-Square Tables, and complie-time computations. Implemented zorbist hashing, transposition table, alpha-beta pruning, and quiscence search. Utilized CUDA for move search when avalible. Integrating NNUE for evaluation.

GitHub Live Demo

Chess NNUE (Efficently Updatable Neural Network)

PyTorch, NumPy, SQL, HDF5

Developing NNUE static evaluation to integrate with my chess engine. Currently reaching 80% accuracy with int_8 quantized model. Working with a dataset of 83 million data points.

GitHub

Natural Language Computing

PyTorch, Transformers, sk-learn, NumPy, pandas

Taking CSC401 at UofT. Applications such as information retrieval, speech recognition and synthesis, machine translation, summarization, and dialogue. N-grams, corpus analysis, neural methods, and information theory.

GitHub Live Demo

Computational Linguistics

PyTorch, NLTK, Transformers, NumPy

Taking CSC485 at UofT. Topics applied in assignments: context-free grammars; chart parsing, statistical parsing; semantics and semantic interpretation; ambiguity resolution techniques; reference resolution.

GitHub Live Demo

Personal Remote ML Server

CUDA, Shell Scripting, Linux, Docker

Reconfigured my retired cryptocurrency mining rig into a remote ML server to train and test my models. Achieved 10-20x faster training times and up to 40x faster in distributed training.