Teaching & Outreach
I teach machine learning the way I do research: a claim only counts once someone has tested it. Everything below is free, open, and built for people who are starting from zero.
ML Academy · mltraining.org
ML Academy is a complete, free interactive curriculum that teaches machine learning, deep learning, and large language models entirely in the browser. There are no videos, no installation, no accounts required to start, and no payment. I designed the pedagogy, wrote all 123 lessons, and implemented every algorithm engine and visualization behind them. The whole project is open source under the MIT license and runs in both English and Turkish.
Why it exists
There are excellent interactive resources for machine learning: TensorFlow Playground, Distill, Seeing Theory, the Poloclub explainers. They share one gap. They show you things, but nothing ever checks whether you learned anything. You drag a slider, think "neat", and move on. ML Academy is built to close exactly that gap, and the mechanism comes straight from my research background in statistical model comparison: other courses show machine learning, this one makes you prove it.
Predict before you see
Before any animation runs, the student commits to an answer. The hit rate becomes a calibration score, which is the one number that shows where intuition is unreliable.
Feel the need before the tool
Nobody is told that significance testing matters. The student builds a model, trusts the number, then re-splits the data with a new random seed and watches the ranking flip. Only then does the test appear.
Prove it by running it
Lessons end with the algorithm itself, not a question about it. The student fills the blanks and presses RUN. Wrong answers are not rejected, they are executed: write a plus instead of a minus and you watch gradient ascent blow the loss up on screen.
No number without a test
Every figure quoted in a lesson is re-derived by a verification script before release. ./dogrula.sh checks the numeric claims, the structure of every lesson, and renders all visualizations against a mock canvas to catch drawing errors.
Curriculum
-
Track 0
Start from zero (15 lessons)
Algorithms, data, what learning is, memorization and overfitting, metrics, data leakage, and statistical evidence.
-
Track 1
Classical machine learning (37 lessons)
k-NN, decision trees, random forests, boosting, logistic regression, SVM, soft decision trees, clustering, PCA and manifold methods.
-
Track 2
Deep learning (22 lessons)
Neurons, backpropagation, optimizers, activations, regularization, batch normalization, CNNs, embeddings, and transfer learning.
-
Track 3
Large language models (29 lessons)
Tokenization, attention, the transformer block, sampling, pre-training and RLHF, hallucination, retrieval-augmented generation, and the KV cache.
-
Track 4
Using AI (20 lessons)
Prompting, evaluation sets and confidence intervals, Elo comparison, debugging a RAG pipeline, agents, LLM-as-judge, red teaming, and cost.
Alongside the lessons there is a catalogue of 25 models. Each entry states what the model does, how it works, when to use it, when not to use it, working code, the key hyperparameters, and the classic pitfall. Every algorithm in the site was written from scratch in plain JavaScript: CART with Gini gain, random forests with bootstrap and feature subsampling, gradient boosting, an MLP with backpropagation, linear SVM, soft decision trees, k-means, PCA with Jacobi eigendecomposition, a BPE tokenizer, skip-gram word2vec with negative sampling, and Elo ratings. No libraries, so every number a student sees is produced live in their own browser.
University Workshops & Instruction
-
2026
Instructor, Data & Analytics Breakout: Explainable Neural Trees
Designed and led the hands-on half of a workshop: students trained their own neural tree, then attacked a deployed model with sensor drift, stuck-at faults, and injected noise to see how an interpretable model fails differently from a black box. The session closed with a contribution sprint in which students opened pull requests against the neural-trees library. All slides, notebooks, and datasets are public.
-
2019–20
Maker Engineer, hands-on STEM curriculum
Led maker-space initiatives and built hands-on engineering curricula, mentoring more than 100 students and organizing community showcase events.
-
2018–19
Educational platform engineering, Technology Transfer Office
Built and tested web-based educational platforms with video streaming and student-tracking systems, which is where the interest in measuring learning rather than delivering content started.
Mentorship
-
2025
Industry Mentor, Penn × Anthropic Sprint Hackathon
-
2025
Mentor, Opportunity Hack / ASU Fall Hackathon
-
2020–
Local Lead & Mentor, NASA Space Apps Challenge
-
2020
National Winner, InovaTIM AI & Robotics High-School Program
Across university programs, maker spaces, hackathons, and national competitions I have taught or mentored more than 500 students. Judging and program-committee service is listed on the Speaking & Service page.
Open Educational Software
-
Web
ML Academy
-
PyPI
-
Web
Courses I Am Prepared to Teach
Introduction to Machine Learning
Supervised and unsupervised learning, model selection, evaluation, and the statistical comparison of models. Undergraduate or first-year graduate.
Deep Learning
Backpropagation, optimization, regularization, convolutional and sequence models, representation learning, and transfer.
Large Language Models and Their Evaluation
Tokenization through transformers, alignment, retrieval-augmented generation, hallucination, evaluation design, and red teaming.
Explainable and Trustworthy AI
Interpretability methods, auditability, safety verification for autonomous systems, and AI governance under the EU AI Act and related standards.