Master of Engineering (Applied ML)

Sentiment Analysis

• Built a pipeline to classify Reddit comments as left-wing, right-wing, positive, negative, or neutral.
• Engineered lexical, syntactic, and affective features, and compared SVM, MLP, and transformer models.
• Designed an evaluation workflow to track precision, recall, and overall accuracy.

Neural Machine Translation

• Implemented key components of a custom Transformer for French-to-English translation using the Hansards dataset.
• Added multi-head attention, beam search decoding, and an evaluation pipeline with perplexity and BLEU scores.
• Benchmarked results against T5, BART, and ChatGPT.

Speech Processing

• Worked on speaker identification and lie detection using audio signals.
• Extracted MFCC features and trained models such as GMMs, GRUs, and DTW-based speaker verification.
• Measured accuracy and word error rates through a clear testing framework.

Overview

Explored classical and modern RL while implementing every algorithm entirely from scratch in Python. Combined theory with projects to understand how methods work under the hood.

What I Did

• Implemented policy evaluation and improvement using dynamic programming for tabular MDPs.
• Applied Monte Carlo, TD(0), TD(λ), and SARSA to environments like CartPole. Visualized learning performance and stability.
• Extended temporal-difference learning with function approximation for continuous state spaces and tested generalization.
• Built a full Deep Q-Network with experience replay, target networks, and epsilon-greedy exploration for high-dimensional tasks.
• Explored human feedback, policy shaping, and RLHF considerations for real-world use.

• Deep Learning (with Prof. Roger Grosse). Explored the math behind neural networks from backpropagation to advanced architectures. Focused on calculus and linear algebra foundations.
• Knowledge Representation and Reasoning. Studied how discrete math is used to model reasoning in AI systems.
• Big Data Analytics. Worked with Microsoft Azure and Databricks on large datasets. Built recommendation systems and applied KNN and SVM.
• AI in Mechatronics. Led a team of three to design a robot that collects dishes and loads a dishwasher. Used YOLO for detection, computer vision for slot finding, and trained a pickup mechanism with reinforcement learning.