Sitemap
A list of all the posts and pages found on the site. For you robots out there, there is an XML version available for digesting as well.
Pages
Posts
24 Hours, 5 Layers, and 0 Caffeine: My Hackathon Dive into Supply Chain Optimization
Published:
A whirlwind 24-hour journey at a university hackathon, where our team attempted to build a full-stack logistics optimizer. I wrestled with MILP and genetic algorithms while my teammates battled UI frameworks, a story of parallel struggles, 4 AM crashes, and what we actually won.
Solving a Million-Step LLM Task with Zero Errors: The Power of Massively Decomposed AI
Published:
A groundbreaking paper introduces MAKER, a system that solves a million-step LLM task with zero errors by breaking problems into tiny pieces and voting on each step. It’s a new path to scalable, reliable AI.
K-Means Clustering from Scratch: Grouping Data with Python
Published:
Uncover the inner workings of K-Means, the fundamental unsupervised learning algorithm, by building it from scratch and watching it group data into meaningful clusters.
Nested Learning: The Illusion of Deep Learning Architectures
Published:
A new paper from Google Research reimagines deep learning as a system of nested optimization problems, revealing how models compress context and opening pathways to more expressive, brain-inspired architectures.
Linear Regression from Scratch: The Foundation of Machine Learning
Published:
Dive into the mathematics and intuition behind Linear Regression, the algorithm that powers prediction, by building it from scratch with gradient descent and feature normalization.
Absolute Zero: AI That Teaches Itself to Reason, With Zero Data
Published:
A groundbreaking new paradigm shows how AI models can teach themselves to reason through self-play, without any human-curated data, and outperform models trained on thousands of expert examples.
Modeling Tennis Matches with Markov Chains
Published:
From the probability of winning a point to the probability of winning a match, exploring how Markov chains model the structure of tennis scoring.
Why Language Models Hallucinate: A Statistical Story Behind Confident Falsehoods
Published:
Large language models don’t just make mistakes, they confidently guess when uncertain. A new paper explains why: our training and evaluation pipelines statistically reward guessing over honesty, making hallucinations an inevitable byproduct of current practices.
Deep Learning vs Machine Learning for Intrusion Detection in Computer Networks
Published:
A new comparative study explores how deep learning stacks up against classic machine learning for intrusion detection in computer networks, revealing both surprising winners and practical trade-offs.
Bypassing AI Text Detectors: What It Really Means for Education
Published:
Simple tricks can fool AI text detectors, but what does that mean for fairness and inclusion in education?
Mastering Minimax: Building a Perfect Tic Tac Toe AI
Published:
An intuitive walkthrough of the Minimax algorithm, how it guarantees optimal play, and how I used it to build a Tic Tac Toe AI.
Random k Conditional Nearest Neighbor for High-Dimensional Data
Published:
RkCNN extends kNN with random feature subsets and separation-based weighting, making nearest-neighbor methods competitive again in high-dimensional classification.
Less is More: Accelerating AI with Advanced Data Strategies
Published:
Insights from the BSC AI Factory talked with NVIDIA researchers on making AI leaner, faster, and more data-savy. A deep dive into why smarter data beats bigger data.
Data Shapley in One Training Run: Measuring Data Contribution at Scale
Published:
A new paper introduces In-Run Data Shapley, a scalable method to measure how much each data point contributes to training a specific model without the prohibitive cost of retraining.
Advanced R-GAN: Smarter GANs for Fraud Detection in Imbalanced Datasets
Published:
A deep dive into Advanced R-GAN, a GAN-based approach for anomaly detection that generates realistic minority samples, improves model performance, and adds explainability to AI systems.
Subliminal Learning in Language Models: Hidden Traits in Benign Data
Published:
A recent paper uncovers a surprising phenomenon: Language Models can transmit hidden behavioral traits to other models through seemingly unrelated data. Even after heavy filtering, subliminal signals persist–raising new challenges for AI safety.
Research Radar: Staying Ahead of the Curve in Scientific Publications
Published:
Keeping up with research is overwhelming. Research Radar helps you track the latest publications and insights effortlessly.
Braking It Down: Dynamic Chunking and the Future of Sequence Modeling
Published:
Instead of processing sequences as one long stream, new research explores dynamic chunking, teaching models to crave tasks into flexible, human-like units.
Climbing the Ladder: Hierarchical Reasoning Models and the Next Step for AI
Published:
Flat chains of though collapse under complexity. Hierarchical reasoning models propose a new path: structuring thought into levels, more like humans do.
The Illusion of Thinking: Why AI Reasoning Models Aren’t Really Thinking
Published:
AI reasoning models write text that looks like thought. But under the surface, the reasoning collapses, revealing that what we’re seeing is not true thinking, it’s an illusion.
An AlphaGo Moment for Neuroevolution? A Paper That Felt Uncannily Familiar
Published:
I recently read a paper that made me do a double take — it felt like a high-powered version of my own Neuroevolutions project. In this post, I break down the key ideas from ‘An AlphaGo Moment for Model Architecture Discovery’ and how they connect to the genetic algorithms I used to evolve RL agents.
How Do Neural Networks Actually Work? (No Math, Just Intuition)
Published:
Neural networks power everything from voice assistants to image recognition, but how do they actually work? In this post, we’ll walk through the core intuition behind neural networks, without the math, and show how these brain-inspired models learn from data.
Data: The Fuel of Machine Learning
Published:
Good models start with good data. In this post, we explore why data quality is often more important than algorithm choice, what makes data ‘dirty,’ and how to clean, prepare, and split your datasets for machine learning success.
Overfitting, Underfitting, and Why Your Model Might Be Lying to You
Published:
Why do machine learning models sometimes perform great on training data but fail miserably on new examples? In this post, we’ll explore the crucial concepts of overfitting and underfitting, and show you how to spot when your model is fooling you into thinking it’s smarter than it really is.
What’s Reinforcement Learning?
Published:
Reinforcement learning is the secret behind how AI learns to play games, drive cars, and control robots, but how does it work? In this post, we’ll break down the key ideas of agents, environments, rewards, and policies to show you how machines learn through trial and error. Ready to level up your AI knowledge? Let’s dive in!
What’s Unsupervised Learning?
Published:
Unsupervised learning is the secret sauce behind many of today’s most powerful AI tools, but how does it find patterns in data without any labels? In this post, we’ll uncover the magic of clustering, association rules, and dimensionality reduction, and show you how these techniques can transform raw data into deep insights. Ready to unlock the mysteries of your data? Let’s dive in!
What’s Supervised Learning?
Published:
Supervised learning is the powerhouse driving modern AI models — but what exactly is it, and how does it turn data into decisions? In this post, we’ll reveal the difference between classification and regression, and show you how these techniques fuel everything from spam filters to stock predictions! Dive in and learn how to make your data work for you!
First Step Into AI: It’s Easier Than You Think
Published:
AI and ML might seem like intimidating worlds of math and code, but they’re far more accessible than most realize. Why Machines Learn demystifies the basics, showing that even the most advanced algorithms build on surprisingly simple ideas. Curious how machines learn? Start here!
portfolio
F1 Race Winner Predictions: From Data Collection to Production-Ready API
Published:
Building an end-to-end machine learning system to predict Formula 1 race winners. From collecting historical race data via FastF1, engineering predictive features, training gradient boosting models, to deploying a FastAPI service—this project showcases the complete ML lifecycle with a focus on production-ready architecture.
Neuroevolutions
Published:
I set out to explore how Genetic Algorithms, inspired by natural selection, can evolve intelligent behavior in classic reinforcement learning environments. Through three increasingly complex challenges, I built and evolved neural networks from scratch using simple yet powerful evolutionary strategies. Wether you’re curious about evolutionary computation, preparing for your next machine learning project, or just watching agents go from clueless to competent, this deep dive is for you.
Building a Multi-Layer Optimization Engine for Livestock Logistics
Published:
Designing and implementing a 5-layer decision support system using MILP, predictive scoring, Monte Carlo simulation, and genetic algorithms to optimize a complex supply chain, increasing profitability by strategically scheduling pig transport.