I'm Anay — a second-year CSE student at IIIT Kottayam who decided semester breaks are for building. What started as "let me learn pandas" turned into a full self-taught roadmap: Python foundations, the entire NumPy/Matplotlib/Seaborn stack, statistics, scikit-learn, SQL, real Kaggle competitions, and now NLP.
My method is simple and a little stubborn: watch a concept once, then type every line of code from scratch. Every topic ends in a project, every portfolio project gets pushed to GitHub with a proper README.
Outside of data science, I run on competitive programming — currently 400+ problems deep on LeetCode via Striver's A2Z sheet, with DP and graphs as my strong suit.
Right now: in Phase 8 — NLP. Built a spam classifier comparing Bag-of-Words and TF-IDF with Naive Bayes and Logistic Regression, explored pretrained word embeddings with GloVe, and just finished an LSTM built on top of a from-scratch embedding layer.
Up next: Phase 9 — MLOps and deployment depth: Docker, FastAPI, and model monitoring, so a model can live somewhere other than a notebook.
Genuinely likes cats. The footer knows this and will not let you forget it.
From Git basics to a deployed fraud-detection model, phase by phase.
Learned Git from scratch, transitioned from Java/OOP knowledge into Python, and started Striver's A2Z DSA sheet with weekend revision.
Python foundations, pandas, NumPy, Matplotlib, Seaborn, government API integration, and statistics — distributions, hypothesis testing, confidence intervals, outlier detection.
scikit-learn end to end — regression, trees, ensembles, clustering, GridSearchCV, pipelines — plus SQL across SQLite and PostgreSQL with real pandas integration.
Titanic, House Prices, and Digit Recognizer submitted; Indian Agriculture, COVID India Dashboard and IPL Analysis built; capstone Customer Churn Prediction with class-imbalance handling and a tuned Random Forest.
XGBoost, Voting and Stacking ensembles, a Keras neural network, a CNN on MNIST (99.09% test accuracy), and a Credit Card Fraud Detection capstone deployed as a Streamlit app.
Tokenization and text cleaning with NLTK, Bag-of-Words vs TF-IDF, a deployed SMS Spam Classifier, pretrained GloVe embeddings, and an LSTM built on a from-scratch embedding layer.
Docker, FastAPI, and model monitoring — taking models further past a single Streamlit script.