All Notes.

Beyond the Click: Slate-Q for Sequential Recommendation Most recommendation systems are designed to maximize immediate engagement—the “next click.” However, true user value is built over entire sessions. In this project, RL-RECSYS, I...

Animating Intelligence: Visualizing AI with Manim-AI Neural networks are often treated as “black boxes,” full of abstract matrices and hidden weights. To bridge the gap between theory and intuition, I developed...

Bonsai-LLM: The 'Small LLMs Lab' Philosophy In an era of trillion-parameter models and massive compute clusters, it’s easy to forget that capacity matters, constraints are the point, and craft beats brute...

Modernizing GPT-2: A 3.1x Throughput Leap with 2025 Optimizations The original GPT-2 architecture, released in 2019, remains the bedrock of modern NLP. However, the “standard” recipe for training Transformers has shifted dramatically. In this...

Causal-Informed Hybrid Online Adaptive Optimization for Ad Load Personalization in Large-Scale Social Networks This paper presents CTRCBO (Cohort-Based Trust Region Contextual Bayesian Optimization), a hybrid framework designed for personalizing ad load in large-scale social networks like Meta.

Generative Retrieval: Scaling PLUM with Hierarchical Semantic IDs The landscape of recommendation systems (RecSys) is undergoing a paradigm shift. We are moving away from traditional “score-and-rank” models toward Generative Retrieval, where an LLM...

Treatment Effect Learning Under Sequential Randomization This research addresses the challenges of causal inference in online experiments (A/B testing) where treatment assignments are sequential.

NxtPost: User to Post Recommendations in Facebook Groups NxtPost is a graph-based recommendation framework designed for Facebook Groups to help users discover relevant content within their communities.

Behavioral Questions In senior technical roles, behavioral interviews are often more about impact, ownership, and cross-functional leadership than technical skill. They test how you handle conflict, complexity,...

ML System Design Designing a machine learning system is vastly different from designing a traditional software system. It requires balancing the needs of high-availability software architecture with the...

Activation Functions Activation functions are the mathematical “gates” that decide whether a neuron should fire or stay dormant. They introduce non-linearity into a neural network, allowing it...

Python Clean Code Python’s philosophy emphasizes readability and simplicity. “Clean code” in Python (often called “Pythonic” code) isn’t just about making it work—it’s about making it expressive and...

Regularization Regularization is a fundamental technique in machine learning used to prevent overfitting by penalizing the complexity of a model. By adding a regularization term to...

Loss functions in machine learning Loss functions are the heartbeat of machine learning models. They quantify the discrepancy between a model’s predictions and the actual ground truth, guiding the optimization...

Failure is not an option For as long as I can remember, I have been captivated by the "Final Frontier." Whether through the lens of a telescope peering into the...

TeTrIS: Template Transformer Networks for Image Segmentation With Shape Priors This research, published in IEEE Transactions on Medical Imaging (TMI), introduces TeTrIS, a network architecture that integrates shape priors into the segmentation process.

Uncertainty in Motion: Probabilistic Robotics from Textbook to Code Probabilistic methods are the foundation of modern robotics. Because sensors are noisy and environments are unpredictable, a robot must be able to represent and reason...

Localization of a Mobile Autonomous Robot Using Extended Kalman Filter This early research focuses on the localization of mobile robots in dynamic environments using Extended Kalman Filters (EKF).