Projects
This is the technical side of the archive: datasets, notebooks, models, dashboards, AI systems, automation projects, tools, experiments, and the discipline of turning messy work into something useful.
Workbench
| Area | What Belongs Here |
|---|---|
| Data Science | Analysis notebooks, dashboards, statistical thinking, visualization, and storytelling with data. |
| Machine Learning | Classification, regression, NLP, recommendation systems, and learning notes. |
| AI Engineering | LLM experiments, agents, prompt systems, model evaluation, automation, and applied AI tools. |
| Experiments | Half-built prototypes, questions worth testing, and small technical discoveries. |
Featured Projects
Library: M-Ana Python Toolkit
A tested, Beta-stage Python toolkit that turns recurring data, AI, modeling, experimentation, forecasting, recommendation, retrieval, and database workflows into reusable package APIs.
Python . Pandas . NumPy . Scikit-learn . SciPy . FAISS . Pinecone . PostgreSQL . SQLAlchemy . PyTorch . TensorFlow . Statsmodels . MkDocs . GitHub Actions
Tool: Phone Video Compressor
A Windows command-line utility that scans an Android phone for large videos, compresses them on a laptop with FFmpeg, and safely replaces the originals through a two-step ADB workflow.
Python . ADB . FFmpeg . ffmpeg-python . PyInstaller . Windows CLI
ML: Lyric Embeddings for Music Recommendations
A lyrics-only music recommender using all-mpnet-base-v2 sentence embeddings and cosine similarity — comparing whole-lyric vs. 4-segment embeddings across 7 test songs from a personal playlist.
Python . Sentence Transformers (all-mpnet-base-v2) . Pandas . Scikit-learn . WordCloud . Matplotlib
ReMo: NLP Movie Recommender System
A hybrid movie recommender system that iterates through four progressively better approaches — metadata cosine similarity, collaborative filtering, NLP embeddings, and a final system combining all three — documented with honest per-attempt ratings and a distinct voice.
Python . Pandas . Scikit-learn . Sentence Transformers (all-MiniLM-L6-v2) . NumPy . Pickle
Observational Data vs. Experimental Data
A simulation-driven reference notebook covering the full distinction between observational and experimental data — confounding, Simpson’s Paradox, A/B test vs. observational bias, and statistical power — each illustrated with live-generated synthetic data.
Python . NumPy . Pandas . SciPy . Scikit-learn . Matplotlib . Seaborn
ML: Evaluating PCA Impact on Churn Prediction Models
A controlled experiment on 790,624 real telecom records — measuring the impact of PCA dimensionality reduction on four classifiers across accuracy, F1, ROC AUC, and training time, with findings that confirm PCA is model-type-dependent.
Python . Pandas . Scikit-learn . XGBoost . MAna (custom package) . Seaborn . Matplotlib
EDA: Traditional vs. Bayesian A/B Testing
A side-by-side implementation of Traditional and Bayesian A/B testing on the Cookie Cats mobile game dataset — two retention metrics, two statistical paradigms, one decision.
Python . Pandas . SciPy . NumPy . Matplotlib . Seaborn
AI: Diagnosify Arabic Medical Symptom Assistant
An Arabic medical-assistant prototype that combines symptom-based disease prediction, conversational symptom extraction, and a RAG-style medical knowledge layer.
Python . Flask . Streamlit . XGBoost . Scikit-learn . Pandas . SQLite . Fuzzy Matching . OpenAI . Pinecone . RAG
Tool: Scroopy-Luu Web Scraper
A Streamlit web-scraping tool that accepts multiple URLs, optionally follows sub-links, extracts page text, logs progress in the UI, and exports scraped content to JSON, Pickle, and CSV.
Python . Streamlit . Requests . BeautifulSoup . Pandas . JSON . Pickle . CSV
AI: Product Conversation Hub
A product-shopping assistant for an industrial catalog, combining web scraping, n8n automation, LLM metadata enrichment, Pinecone vector search, and a Streamlit chat interface.
Python . Streamlit . n8n . OpenAI API . Anthropic Claude . Pinecone . Google Drive API . Google Sheets API . BeautifulSoup . Requests . PyMuPDF . Docker . Google App Engine . JSON . RAG
Tool: RNN Time-Series Forecasting Gradio UI
A Gradio interface for training PyTorch RNN, LSTM, or GRU forecasting models on uploaded time-series CSV files, then returning MAE, downloadable model weights, prediction plots, and future forecasts.
Python . PyTorch . Gradio . pandas . scikit-learn . Matplotlib . RNN . LSTM . GRU
Stats: Star Power and Movie Success
A statistical and machine-learning study on TMDB movie data, testing whether star actors actually improve revenue, popularity, audience engagement, and profitability.
Python . Pandas . NumPy . SciPy . Seaborn . Plotly . Scikit-learn . XGBoost . Bayesian Testing . A/B Testing
EDA + ML: Music Trends Analysis & ANN Trendiness Prediction
A Spotify 2024 streaming dataset EDA with a hand-engineered TikTok trendiness score, followed by a PyTorch ANN classifier with early stopping and gradient clipping that predicts a song’s trendiness category from cross-platform streaming metrics.
Python . PyTorch . Pandas . Scikit-learn . Plotly . Seaborn
AI: LLM CSV Agent
A local LLM-powered CSV agent that uses LangChain and Ollama to answer analytical questions over tabular data, turning natural-language prompts into pandas-style reasoning.
Python . LangChain . LangChain Experimental . Ollama . DeepSeek Coder . Pandas . CSV
AI: Fully Open-Source RAG System
A fully open-source Retrieval-Augmented Generation system that lets users upload a PDF, embeds its content with Sentence Transformers, retrieves relevant chunks with FAISS, and answers questions through a local Ollama/Llama 3 workflow.
Python . LangChain . Ollama . Llama 3 . Sentence Transformers . FAISS . Streamlit . PyPDFLoader
Tool: LangFlow Dockerized
A lightweight Docker setup for running LangFlow as a reproducible visual environment for building and testing LLM, agent, and RAG workflows.
Docker . Python . LangFlow . LangChain . LLM Workflows
AI: Doctor Intelligent Graduation Project
A graduation capstone project for an AI-powered mental-health knowledge assistant, combining a classic educational web section, bilingual chat endpoints, GPT-based responses, RAG retrieval with Pinecone, and a PHP frontend.
Python . Flask . LangChain . OpenAI GPT Models . Pinecone . Sentence Transformers . RAG . PHP . JavaScript . HTML/CSS
AI: PDF Chatbot with Streamlit and OpenAI
An early RAG-style PDF chatbot that extracts pages from a document, builds a vector index, and lets users ask questions through a Streamlit chat interface powered by OpenAI.
Python . Streamlit . OpenAI . GPT-3.5 . LlamaIndex . LangChain . PyPDF2 . JSON
AI: IMDB Sentiment Classification with Fine-Tuned BERT
Fine-tuning bert-base-uncased on 10,000 IMDB movie reviews for binary sentiment classification — including a custom Dataset class, a linear warmup scheduler, and a reusable inference function.
Python . PyTorch . Hugging Face Transformers . Scikit-learn . Pandas
ML: Australian Horse Racing Rating System
A horse-racing analytics pipeline that collects Australian race data, engineers form and rating features, combines XGBoost and neural-network predictions, and exports daily rating sheets.
Python . Pandas . NumPy . XGBoost . TensorFlow . Keras . Pickle . CSV . Excel
Tool: Dockerized PyTorch CNN Classifier API
A Dockerized Flask API that serves a trained PyTorch CNN for image classification, turning a notebook-trained model into a local inference service.
Python . PyTorch . torchvision . Flask . Docker . Pillow . CNN
AI: Azure OpenAI Document Processing and RAG System
An Azure OpenAI prototype for document summarization, tagging, classification, and RAG-style question answering over service workflow and vehicle-license data.
Python . Azure OpenAI . OpenAI API . GPT-4 . Pinecone . Sentence Transformers . Gradio . PyPDF2 . JSON . RAG
Tool: Flask Web Scraper API
A Flask-based scraping and ingestion API that collects web pages, PDFs, and Google Drive documents, cleans them into structured JSON, and stores the results for downstream search or AI workflows.
Python . Flask . BeautifulSoup . Requests . PyPDF2 . MongoDB . Google Drive API . Langflow . JSON . CSV
EDA: IMDB / TMDb — Deep Movie Analysis
A comprehensive deep-dive into ~5,000 TMDb movies — systematically profiling budget, revenue, profit, runtime, genres, keywords, cast, directors, production companies, and release timing across every major performance metric.
Python . Pandas . Plotly . Seaborn . Matplotlib . WordCloud
AI: PyTorch Insects Image Classification
A binary insect image classifier built in PyTorch from scratch — featuring a custom Dataset class, manual shape-tracing to compute layer dimensions, and a full train/eval loop with loss curve visualization.
Python . PyTorch . torchvision . Scikit-learn . Pandas . Matplotlib
AI: ASL Recognition with TinyVGG & Custom CNN (100% Accuracy)
Two PyTorch CNNs trained to classify 29 American Sign Language hand gestures from 87,000 images — TinyVGG achieves 96.5% accuracy; a deeper custom CNN reaches 100%.
Python . PyTorch . torchvision . torchinfo
ML: Simple Neural Network with PyTorch
A from-scratch PyTorch feedforward neural network for binary tabular classification — a focused exercise in building and training a custom nn.Module by hand.
Python . PyTorch . Pandas . Scikit-learn . Matplotlib
EDA: Boston Property Tableau Dashboard
A Tableau dashboard analyzing Boston condominium assessment data, using cleaned property records to compare city value patterns, construction periods, and features that influence condo prices.
Tableau . Excel . CSV . Data Cleaning . Data Visualization . Dashboard Design
ML: Customer Churn Prediction for Telecom Dataset
A telecom churn prediction project that labels customer-week records from disconnected-number data, preprocesses usage and revenue features, balances the classes, compares LR/SVC/RF/XGBoost/LSTM models, and exports trained models, feature importances, predictions, and Power BI artifacts.
Python . Pandas . Scikit-learn . XGBoost . Random Forest . SVC . Logistic Regression . LSTM . ELI5 . Power BI
ML: Futures Trading Analytics Model
A market microstructure experiment on ES futures block data, testing whether volume-profile and breakout features could predict directional price movement.
Python . Pandas . Scikit-learn . XGBoost . KNN . Decision Tree . Random Forest . SVC . LSTM
AI: LLM-Powered Product Recommendation System (davinci-instruct-beta)
Three prompt engineering experiments testing whether OpenAI’s davinci-instruct-beta can act as a product recommendation system on real e-commerce data from Bobcat in a Box — with an honest verdict: it can’t, and here’s exactly why.
Python . OpenAI API (davinci-instruct-beta) . LlamaIndex . LangChain . Pandas . Scikit-learn
ML: Predictive Analytics for Work Orders
A large operational forecasting project that predicts work-order volume, facility demand, location and escort risk, and duration bins from work-order, detainee, and census data using XGBoost, linear models, and deployment-ready outputs.
Python . Pandas . XGBoost . Linear Regression . Hybrid Modeling . Scikit-learn . Joblib . Power BI . Parquet . Excel
AI: Vessel Object Detection with YOLOv5
Training and evaluating a YOLOv5s model to detect 10 classes of maritime vessels and safety equipment in real-world waterway images, achieving 0.861 mAP@0.5 overall with a structured post-processing pipeline to evaluate predictions against ground truth labels.
Python . YOLOv5 (Ultralytics) . PyTorch . Pandas . Scikit-learn . Matplotlib
ML: Customer Event Prediction for Subscription Retention
A client-style machine learning project for a food-box subscription business, predicting each customer’s next subscription event and timing from historical event logs using a prepared tabular sequence pipeline and TensorFlow LSTM model.
Python . Pandas . TensorFlow . Keras . LSTM . Scikit-learn . ELI5 . MAna . Plotly . Seaborn
ML: Solar Energy Prediction with LSTM, ELM, and Ensemble Models
A solar energy forecasting project that preprocesses lagged solar time-series data, trains tuned LSTM and ELM neural models, combines them into an ensemble, and saves models, scalers, predictions, and plotting data for reuse.
Python . Pandas . TensorFlow . Keras . LSTM . ELM . Ensemble Modeling . Bayesian Optimization . Scikit-learn . Statsmodels . Plotly
ML: Incidents Prediction with XGBoost
A machine learning project that forecasts daily incident volume and predicts incident severity levels from joined incident, detainee, and profile datasets using XGBoost models and exported Excel prediction reports.
Python . Pandas . XGBoost . Scikit-learn . GridSearchCV . ELI5 . Statsmodels . Plotly . Seaborn . Parquet . Excel
ML: Solar Irradiance Tracking with Time Series Forecasting
A time-series forecasting project that predicts next-hour solar irradiance from hourly weather and irradiance data, comparing LSTM, bidirectional LSTM, KNN, SVR, and ARIMA models with saved model artifacts.
Python . Pandas . LSTM . Bidirectional LSTM . KNN . SVR . ARIMA . Scikit-learn . Keras . Statsmodels . GridSearchCV
ML: Text Classification for Women's Clothing Reviews
A systematic XGBoost classification experiment on 23,000+ clothing reviews — comparing five configurations across feature type (non-textual, textual, combined) and class balancing strategy (none, under-sampling, over-sampling).
Python . Pandas . XGBoost . Scikit-learn . NLTK . Plotly . Seaborn
EDA: Google Play Store — Data Wrangling
A systematic data wrangling notebook that produces a publication-ready clean version of the Google Play Store dataset — handling six messy columns, a tiered Rating imputation strategy, duplicate resolution, and genre normalization.
Python . Pandas . MAna (custom package) . Matplotlib
EDA + ML: Store Sales Time Series Forecasting
A multi-dataset time series EDA and forecasting entry for Kaggle’s Store Sales competition — joining five data sources, profiling oil prices and transactions for seasonality, and forecasting Ecuadorian grocery sales with a Linear Regression baseline.
Python . Pandas . Scikit-learn . Plotly . Seaborn . SciPy
ML: Goodreads Book Search Engine
A lightweight machine learning search engine for Goodreads books that cleans book titles, vectorizes them with TF-IDF, retrieves similar titles with cosine similarity, and ranks results by a custom popularity score.
Python . Pandas . Scikit-learn . TF-IDF . Cosine Similarity . Regex . Seaborn
AI: Deep RNN (LSTM) Text Generation on Moby Dick
Training a three-layer stacked LSTM on the first quarter of Moby Dick to build a character-level next-word text generator — the first sequence modeling project in the portfolio.
Python . TensorFlow / Keras . spaCy . NumPy . Pandas
ML: Alumni Data Analysis and Donor Clustering
A client-style alumni advancement analysis that cleans a large Excel dataset, answers faculty and regional targeting questions, clusters likely donors with K-Means, and exports ranked donor lists for outreach.
Python . Pandas . Scikit-learn . K-Means . StandardScaler . Silhouette Score . Plotly . Seaborn . Excel
EDA: Goodreads — What Makes a Perfect Book?
21 questions answered across 11,000+ Goodreads books — from top authors and publishers to a custom popularity score that combines average rating and read count into a single fair ranking metric.
Python . Pandas . Seaborn . Matplotlib
ML: Email Spam Classification & Anomaly Detection
Four supervised classifiers benchmarked on 5,500+ emails with TF-IDF features, then a Local Outlier Factor anomaly detector applied to the same problem — correctly labeling 81% of emails with zero label supervision.
Python . Pandas . Scikit-learn . XGBoost . NLTK / TF-IDF . Plotly . Seaborn
EDA: Streaming Platforms Tableau Dashboard
An early Tableau dashboard comparing streaming-platform catalogs after reshaping Netflix, Disney+, and Amazon title data into analysis-ready tables.
Tableau . Python . Pandas . CSV . Data Modeling . Data Visualization
EDA: YouTube Trending Videos Tableau Dashboard
An early Tableau dashboard analyzing YouTube trending videos by views, tags, categories, channels, states, likes, dislikes, and comments.
Tableau . Python . Pandas . CSV . Data Visualization . Dashboard Design
EDA: Chinook Music Market Tableau Dashboard
An early Tableau dashboard built on the Chinook music database, exploring sales by country, genre, artist, and revenue.
Tableau . SQL . SQLite . Excel . Data Visualization . Dashboard Design
ML: Credit Card Fraud Detection
A fraud detection classifier on 284,000+ credit card transactions — tackling extreme class imbalance (0.17% fraud) through three progressively stable under-sampling strategies before landing on a feature-selected SVC with zero false negatives.
Python . Pandas . Scikit-learn . eli5 . Plotly . Seaborn
EDA + ML: Customer Personality Analysis & Segmentation
A full-pipeline analysis of a marketing dataset — EDA surfacing income, spending, and complaint patterns across customer demographics, followed by K-Means clustering validated with both the elbow method and silhouette scores.
Python . Pandas . Scikit-learn . Plotly . Seaborn
EDA + ML: Google Play Store App Analysis & Rating Prediction
A dual-dataset EDA of 10,000+ Google Play apps and their user reviews — with a custom composite popularity score, sentiment-app feature cross-analysis, and a three-model regression benchmark for predicting app ratings.
Python . Pandas . Scikit-learn . XGBoost . eli5 . Plotly . Seaborn
ML: Loan Default Prediction
This is a Kaggle competition notebook that builds a machine learning classifier to predict loan default outcomes (i.e., whether a loan will result in a loss, and how much). It was submitted to the Loan Default Prediction competition on Kaggle.
Python . Pandas . Scikit-learn
EDA: Netflix Movies & TV Shows
A fast EDA of Netflix’s catalog — seven questions answered across content type, release trends, ratings, countries, cast, directors, and genres.
Python . Pandas . Plotly . Seaborn
ML: Twitter Stock Price Forecasting with ARIMA, SARIMA & NeuralProphet
A rigorous time series analysis of Twitter’s stock price — stationarity testing, ACF/PACF-guided ARIMA order selection, SARIMA comparison, and an honest NeuralProphet failure documented and explained.
Python . Pandas . Statsmodels . NeuralProphet . SciPy . Plotly . Seaborn
EDA: Students Performance in Exams
An exploratory analysis of 1,000 student exam records, showing how demographics, lunch type, parental education, and test preparation relate to math, reading, and writing scores.
Python . Pandas . Plotly . Seaborn . Matplotlib
ML: Simple or Complex Models?
A deliberate experiment on the Acoustic Extinguisher Fire dataset — comparing Decision Tree, Random Forest, and SVC to test whether model complexity correlates with accuracy.
Python . Pandas . Scikit-learn . Plotly . Seaborn
EDA: Video Game Sales — High-Level Visualizations
A high-level exploratory analysis of video game releases and sales across platforms, genres, publishers, years, and regions, with reusable visualization functions.
Python . Pandas . Plotly . Seaborn . Matplotlib . pandas-profiling
ML: US Accidents Severity & Count Prediction
Two ML models on 3M+ US accident records — a KNN classifier for accident severity and an XGBoost regressor for daily accident count — with a companion EDA notebook and a frank mid-notebook pivot when the first approach underperformed.
Python . Pandas . Scikit-learn . XGBoost . Seaborn . Matplotlib
ML: Mall Customer Segmentation with K-Means Clustering
An unsupervised learning project that segments mall customers into five interpretable groups using age, income, spending score, K-Means, and 3D visualization.
Python . Pandas . Scikit-learn . Plotly . Seaborn
ML: E-Commerce Sales Forecasting & Customer Clustering
A dual-objective ML project on a UK retail dataset — forecasting daily order quantities with three regression models and segmenting customers into behavioral personas using K-Means clustering.
Python . Pandas . Scikit-learn . CatBoost . Plotly . Seaborn
ML: Daily Climate Time Series Forecasting
A forecasting project for Delhi climate variables comparing date-feature KNN baselines, full-feature KNN models, and NeuralProphet time-series forecasts.
Python . Pandas . Scikit-learn . NeuralProphet . Plotly . Seaborn
ML: House Prices — Advanced Regression Techniques
My first ML project — a four-model regression benchmark on the Kaggle House Prices dataset, built one month into learning machine learning.
Python . Pandas . Scikit-learn . XGBoost . Seaborn
EDA: E-Commerce — eBay, Flipkart & Pakistan Market Analysis
A cross-platform analysis of eBay, Flipkart, and Pakistan’s e-commerce market to measure the impact of digital commerce on sales volume, pricing, customer satisfaction, and seasonal trends.
Python . Pandas . Matplotlib . Seaborn
AI: Diagnosify Arabic Medical Symptom Assistant
An Arabic medical-assistant prototype that combines symptom-based disease prediction, conversational symptom extraction, and a RAG-style medical knowledge layer.
Python . Flask . Streamlit . XGBoost . Scikit-learn . Pandas . SQLite . Fuzzy Matching . OpenAI . Pinecone . RAG
AI: Product Conversation Hub
A product-shopping assistant for an industrial catalog, combining web scraping, n8n automation, LLM metadata enrichment, Pinecone vector search, and a Streamlit chat interface.
Python . Streamlit . n8n . OpenAI API . Anthropic Claude . Pinecone . Google Drive API . Google Sheets API . BeautifulSoup . Requests . PyMuPDF . Docker . Google App Engine . JSON . RAG
AI: LLM CSV Agent
A local LLM-powered CSV agent that uses LangChain and Ollama to answer analytical questions over tabular data, turning natural-language prompts into pandas-style reasoning.
Python . LangChain . LangChain Experimental . Ollama . DeepSeek Coder . Pandas . CSV
AI: Fully Open-Source RAG System
A fully open-source Retrieval-Augmented Generation system that lets users upload a PDF, embeds its content with Sentence Transformers, retrieves relevant chunks with FAISS, and answers questions through a local Ollama/Llama 3 workflow.
Python . LangChain . Ollama . Llama 3 . Sentence Transformers . FAISS . Streamlit . PyPDFLoader
AI: Doctor Intelligent Graduation Project
A graduation capstone project for an AI-powered mental-health knowledge assistant, combining a classic educational web section, bilingual chat endpoints, GPT-based responses, RAG retrieval with Pinecone, and a PHP frontend.
Python . Flask . LangChain . OpenAI GPT Models . Pinecone . Sentence Transformers . RAG . PHP . JavaScript . HTML/CSS
AI: PDF Chatbot with Streamlit and OpenAI
An early RAG-style PDF chatbot that extracts pages from a document, builds a vector index, and lets users ask questions through a Streamlit chat interface powered by OpenAI.
Python . Streamlit . OpenAI . GPT-3.5 . LlamaIndex . LangChain . PyPDF2 . JSON
AI: IMDB Sentiment Classification with Fine-Tuned BERT
Fine-tuning bert-base-uncased on 10,000 IMDB movie reviews for binary sentiment classification — including a custom Dataset class, a linear warmup scheduler, and a reusable inference function.
Python . PyTorch . Hugging Face Transformers . Scikit-learn . Pandas
AI: Azure OpenAI Document Processing and RAG System
An Azure OpenAI prototype for document summarization, tagging, classification, and RAG-style question answering over service workflow and vehicle-license data.
Python . Azure OpenAI . OpenAI API . GPT-4 . Pinecone . Sentence Transformers . Gradio . PyPDF2 . JSON . RAG
AI: PyTorch Insects Image Classification
A binary insect image classifier built in PyTorch from scratch — featuring a custom Dataset class, manual shape-tracing to compute layer dimensions, and a full train/eval loop with loss curve visualization.
Python . PyTorch . torchvision . Scikit-learn . Pandas . Matplotlib
AI: ASL Recognition with TinyVGG & Custom CNN (100% Accuracy)
Two PyTorch CNNs trained to classify 29 American Sign Language hand gestures from 87,000 images — TinyVGG achieves 96.5% accuracy; a deeper custom CNN reaches 100%.
Python . PyTorch . torchvision . torchinfo
AI: LLM-Powered Product Recommendation System (davinci-instruct-beta)
Three prompt engineering experiments testing whether OpenAI’s davinci-instruct-beta can act as a product recommendation system on real e-commerce data from Bobcat in a Box — with an honest verdict: it can’t, and here’s exactly why.
Python . OpenAI API (davinci-instruct-beta) . LlamaIndex . LangChain . Pandas . Scikit-learn
AI: Vessel Object Detection with YOLOv5
Training and evaluating a YOLOv5s model to detect 10 classes of maritime vessels and safety equipment in real-world waterway images, achieving 0.861 mAP@0.5 overall with a structured post-processing pipeline to evaluate predictions against ground truth labels.
Python . YOLOv5 (Ultralytics) . PyTorch . Pandas . Scikit-learn . Matplotlib
AI: Deep RNN (LSTM) Text Generation on Moby Dick
Training a three-layer stacked LSTM on the first quarter of Moby Dick to build a character-level next-word text generator — the first sequence modeling project in the portfolio.
Python . TensorFlow / Keras . spaCy . NumPy . Pandas
ML: Lyric Embeddings for Music Recommendations
A lyrics-only music recommender using all-mpnet-base-v2 sentence embeddings and cosine similarity — comparing whole-lyric vs. 4-segment embeddings across 7 test songs from a personal playlist.
Python . Sentence Transformers (all-mpnet-base-v2) . Pandas . Scikit-learn . WordCloud . Matplotlib
ReMo: NLP Movie Recommender System
A hybrid movie recommender system that iterates through four progressively better approaches — metadata cosine similarity, collaborative filtering, NLP embeddings, and a final system combining all three — documented with honest per-attempt ratings and a distinct voice.
Python . Pandas . Scikit-learn . Sentence Transformers (all-MiniLM-L6-v2) . NumPy . Pickle
ML: Evaluating PCA Impact on Churn Prediction Models
A controlled experiment on 790,624 real telecom records — measuring the impact of PCA dimensionality reduction on four classifiers across accuracy, F1, ROC AUC, and training time, with findings that confirm PCA is model-type-dependent.
Python . Pandas . Scikit-learn . XGBoost . MAna (custom package) . Seaborn . Matplotlib
EDA + ML: Music Trends Analysis & ANN Trendiness Prediction
A Spotify 2024 streaming dataset EDA with a hand-engineered TikTok trendiness score, followed by a PyTorch ANN classifier with early stopping and gradient clipping that predicts a song’s trendiness category from cross-platform streaming metrics.
Python . PyTorch . Pandas . Scikit-learn . Plotly . Seaborn
ML: Australian Horse Racing Rating System
A horse-racing analytics pipeline that collects Australian race data, engineers form and rating features, combines XGBoost and neural-network predictions, and exports daily rating sheets.
Python . Pandas . NumPy . XGBoost . TensorFlow . Keras . Pickle . CSV . Excel
ML: Simple Neural Network with PyTorch
A from-scratch PyTorch feedforward neural network for binary tabular classification — a focused exercise in building and training a custom nn.Module by hand.
Python . PyTorch . Pandas . Scikit-learn . Matplotlib
ML: Customer Churn Prediction for Telecom Dataset
A telecom churn prediction project that labels customer-week records from disconnected-number data, preprocesses usage and revenue features, balances the classes, compares LR/SVC/RF/XGBoost/LSTM models, and exports trained models, feature importances, predictions, and Power BI artifacts.
Python . Pandas . Scikit-learn . XGBoost . Random Forest . SVC . Logistic Regression . LSTM . ELI5 . Power BI
ML: Futures Trading Analytics Model
A market microstructure experiment on ES futures block data, testing whether volume-profile and breakout features could predict directional price movement.
Python . Pandas . Scikit-learn . XGBoost . KNN . Decision Tree . Random Forest . SVC . LSTM
ML: Predictive Analytics for Work Orders
A large operational forecasting project that predicts work-order volume, facility demand, location and escort risk, and duration bins from work-order, detainee, and census data using XGBoost, linear models, and deployment-ready outputs.
Python . Pandas . XGBoost . Linear Regression . Hybrid Modeling . Scikit-learn . Joblib . Power BI . Parquet . Excel
ML: Customer Event Prediction for Subscription Retention
A client-style machine learning project for a food-box subscription business, predicting each customer’s next subscription event and timing from historical event logs using a prepared tabular sequence pipeline and TensorFlow LSTM model.
Python . Pandas . TensorFlow . Keras . LSTM . Scikit-learn . ELI5 . MAna . Plotly . Seaborn
ML: Solar Energy Prediction with LSTM, ELM, and Ensemble Models
A solar energy forecasting project that preprocesses lagged solar time-series data, trains tuned LSTM and ELM neural models, combines them into an ensemble, and saves models, scalers, predictions, and plotting data for reuse.
Python . Pandas . TensorFlow . Keras . LSTM . ELM . Ensemble Modeling . Bayesian Optimization . Scikit-learn . Statsmodels . Plotly
ML: Incidents Prediction with XGBoost
A machine learning project that forecasts daily incident volume and predicts incident severity levels from joined incident, detainee, and profile datasets using XGBoost models and exported Excel prediction reports.
Python . Pandas . XGBoost . Scikit-learn . GridSearchCV . ELI5 . Statsmodels . Plotly . Seaborn . Parquet . Excel
ML: Solar Irradiance Tracking with Time Series Forecasting
A time-series forecasting project that predicts next-hour solar irradiance from hourly weather and irradiance data, comparing LSTM, bidirectional LSTM, KNN, SVR, and ARIMA models with saved model artifacts.
Python . Pandas . LSTM . Bidirectional LSTM . KNN . SVR . ARIMA . Scikit-learn . Keras . Statsmodels . GridSearchCV
ML: Text Classification for Women's Clothing Reviews
A systematic XGBoost classification experiment on 23,000+ clothing reviews — comparing five configurations across feature type (non-textual, textual, combined) and class balancing strategy (none, under-sampling, over-sampling).
Python . Pandas . XGBoost . Scikit-learn . NLTK . Plotly . Seaborn
EDA + ML: Store Sales Time Series Forecasting
A multi-dataset time series EDA and forecasting entry for Kaggle’s Store Sales competition — joining five data sources, profiling oil prices and transactions for seasonality, and forecasting Ecuadorian grocery sales with a Linear Regression baseline.
Python . Pandas . Scikit-learn . Plotly . Seaborn . SciPy
ML: Goodreads Book Search Engine
A lightweight machine learning search engine for Goodreads books that cleans book titles, vectorizes them with TF-IDF, retrieves similar titles with cosine similarity, and ranks results by a custom popularity score.
Python . Pandas . Scikit-learn . TF-IDF . Cosine Similarity . Regex . Seaborn
ML: Alumni Data Analysis and Donor Clustering
A client-style alumni advancement analysis that cleans a large Excel dataset, answers faculty and regional targeting questions, clusters likely donors with K-Means, and exports ranked donor lists for outreach.
Python . Pandas . Scikit-learn . K-Means . StandardScaler . Silhouette Score . Plotly . Seaborn . Excel
ML: Email Spam Classification & Anomaly Detection
Four supervised classifiers benchmarked on 5,500+ emails with TF-IDF features, then a Local Outlier Factor anomaly detector applied to the same problem — correctly labeling 81% of emails with zero label supervision.
Python . Pandas . Scikit-learn . XGBoost . NLTK / TF-IDF . Plotly . Seaborn
ML: Credit Card Fraud Detection
A fraud detection classifier on 284,000+ credit card transactions — tackling extreme class imbalance (0.17% fraud) through three progressively stable under-sampling strategies before landing on a feature-selected SVC with zero false negatives.
Python . Pandas . Scikit-learn . eli5 . Plotly . Seaborn
EDA + ML: Customer Personality Analysis & Segmentation
A full-pipeline analysis of a marketing dataset — EDA surfacing income, spending, and complaint patterns across customer demographics, followed by K-Means clustering validated with both the elbow method and silhouette scores.
Python . Pandas . Scikit-learn . Plotly . Seaborn
EDA + ML: Google Play Store App Analysis & Rating Prediction
A dual-dataset EDA of 10,000+ Google Play apps and their user reviews — with a custom composite popularity score, sentiment-app feature cross-analysis, and a three-model regression benchmark for predicting app ratings.
Python . Pandas . Scikit-learn . XGBoost . eli5 . Plotly . Seaborn
ML: Loan Default Prediction
This is a Kaggle competition notebook that builds a machine learning classifier to predict loan default outcomes (i.e., whether a loan will result in a loss, and how much). It was submitted to the Loan Default Prediction competition on Kaggle.
Python . Pandas . Scikit-learn
ML: Twitter Stock Price Forecasting with ARIMA, SARIMA & NeuralProphet
A rigorous time series analysis of Twitter’s stock price — stationarity testing, ACF/PACF-guided ARIMA order selection, SARIMA comparison, and an honest NeuralProphet failure documented and explained.
Python . Pandas . Statsmodels . NeuralProphet . SciPy . Plotly . Seaborn
ML: Simple or Complex Models?
A deliberate experiment on the Acoustic Extinguisher Fire dataset — comparing Decision Tree, Random Forest, and SVC to test whether model complexity correlates with accuracy.
Python . Pandas . Scikit-learn . Plotly . Seaborn
ML: US Accidents Severity & Count Prediction
Two ML models on 3M+ US accident records — a KNN classifier for accident severity and an XGBoost regressor for daily accident count — with a companion EDA notebook and a frank mid-notebook pivot when the first approach underperformed.
Python . Pandas . Scikit-learn . XGBoost . Seaborn . Matplotlib
ML: Mall Customer Segmentation with K-Means Clustering
An unsupervised learning project that segments mall customers into five interpretable groups using age, income, spending score, K-Means, and 3D visualization.
Python . Pandas . Scikit-learn . Plotly . Seaborn
ML: E-Commerce Sales Forecasting & Customer Clustering
A dual-objective ML project on a UK retail dataset — forecasting daily order quantities with three regression models and segmenting customers into behavioral personas using K-Means clustering.
Python . Pandas . Scikit-learn . CatBoost . Plotly . Seaborn
ML: Daily Climate Time Series Forecasting
A forecasting project for Delhi climate variables comparing date-feature KNN baselines, full-feature KNN models, and NeuralProphet time-series forecasts.
Python . Pandas . Scikit-learn . NeuralProphet . Plotly . Seaborn
ML: House Prices — Advanced Regression Techniques
My first ML project — a four-model regression benchmark on the Kaggle House Prices dataset, built one month into learning machine learning.
Python . Pandas . Scikit-learn . XGBoost . Seaborn
EDA: IMDB / TMDb — Deep Movie Analysis
A comprehensive deep-dive into ~5,000 TMDb movies — systematically profiling budget, revenue, profit, runtime, genres, keywords, cast, directors, production companies, and release timing across every major performance metric.
Python . Pandas . Plotly . Seaborn . Matplotlib . WordCloud
EDA: Boston Property Tableau Dashboard
A Tableau dashboard analyzing Boston condominium assessment data, using cleaned property records to compare city value patterns, construction periods, and features that influence condo prices.
Tableau . Excel . CSV . Data Cleaning . Data Visualization . Dashboard Design
EDA: Google Play Store — Data Wrangling
A systematic data wrangling notebook that produces a publication-ready clean version of the Google Play Store dataset — handling six messy columns, a tiered Rating imputation strategy, duplicate resolution, and genre normalization.
Python . Pandas . MAna (custom package) . Matplotlib
EDA: Goodreads — What Makes a Perfect Book?
21 questions answered across 11,000+ Goodreads books — from top authors and publishers to a custom popularity score that combines average rating and read count into a single fair ranking metric.
Python . Pandas . Seaborn . Matplotlib
EDA: Streaming Platforms Tableau Dashboard
An early Tableau dashboard comparing streaming-platform catalogs after reshaping Netflix, Disney+, and Amazon title data into analysis-ready tables.
Tableau . Python . Pandas . CSV . Data Modeling . Data Visualization
EDA: YouTube Trending Videos Tableau Dashboard
An early Tableau dashboard analyzing YouTube trending videos by views, tags, categories, channels, states, likes, dislikes, and comments.
Tableau . Python . Pandas . CSV . Data Visualization . Dashboard Design
EDA: Chinook Music Market Tableau Dashboard
An early Tableau dashboard built on the Chinook music database, exploring sales by country, genre, artist, and revenue.
Tableau . SQL . SQLite . Excel . Data Visualization . Dashboard Design
EDA: Netflix Movies & TV Shows
A fast EDA of Netflix’s catalog — seven questions answered across content type, release trends, ratings, countries, cast, directors, and genres.
Python . Pandas . Plotly . Seaborn
EDA: Students Performance in Exams
An exploratory analysis of 1,000 student exam records, showing how demographics, lunch type, parental education, and test preparation relate to math, reading, and writing scores.
Python . Pandas . Plotly . Seaborn . Matplotlib
EDA: Video Game Sales — High-Level Visualizations
A high-level exploratory analysis of video game releases and sales across platforms, genres, publishers, years, and regions, with reusable visualization functions.
Python . Pandas . Plotly . Seaborn . Matplotlib . pandas-profiling
EDA: E-Commerce — eBay, Flipkart & Pakistan Market Analysis
A cross-platform analysis of eBay, Flipkart, and Pakistan’s e-commerce market to measure the impact of digital commerce on sales volume, pricing, customer satisfaction, and seasonal trends.
Python . Pandas . Matplotlib . Seaborn
Observational Data vs. Experimental Data
A simulation-driven reference notebook covering the full distinction between observational and experimental data — confounding, Simpson’s Paradox, A/B test vs. observational bias, and statistical power — each illustrated with live-generated synthetic data.
Python . NumPy . Pandas . SciPy . Scikit-learn . Matplotlib . Seaborn
EDA: Traditional vs. Bayesian A/B Testing
A side-by-side implementation of Traditional and Bayesian A/B testing on the Cookie Cats mobile game dataset — two retention metrics, two statistical paradigms, one decision.
Python . Pandas . SciPy . NumPy . Matplotlib . Seaborn
Stats: Star Power and Movie Success
A statistical and machine-learning study on TMDB movie data, testing whether star actors actually improve revenue, popularity, audience engagement, and profitability.
Python . Pandas . NumPy . SciPy . Seaborn . Plotly . Scikit-learn . XGBoost . Bayesian Testing . A/B Testing
Library: M-Ana Python Toolkit
A tested, Beta-stage Python toolkit that turns recurring data, AI, modeling, experimentation, forecasting, recommendation, retrieval, and database workflows into reusable package APIs.
Python . Pandas . NumPy . Scikit-learn . SciPy . FAISS . Pinecone . PostgreSQL . SQLAlchemy . PyTorch . TensorFlow . Statsmodels . MkDocs . GitHub Actions
Tool: Phone Video Compressor
A Windows command-line utility that scans an Android phone for large videos, compresses them on a laptop with FFmpeg, and safely replaces the originals through a two-step ADB workflow.
Python . ADB . FFmpeg . ffmpeg-python . PyInstaller . Windows CLI
Tool: Scroopy-Luu Web Scraper
A Streamlit web-scraping tool that accepts multiple URLs, optionally follows sub-links, extracts page text, logs progress in the UI, and exports scraped content to JSON, Pickle, and CSV.
Python . Streamlit . Requests . BeautifulSoup . Pandas . JSON . Pickle . CSV
Tool: RNN Time-Series Forecasting Gradio UI
A Gradio interface for training PyTorch RNN, LSTM, or GRU forecasting models on uploaded time-series CSV files, then returning MAE, downloadable model weights, prediction plots, and future forecasts.
Python . PyTorch . Gradio . pandas . scikit-learn . Matplotlib . RNN . LSTM . GRU
Tool: LangFlow Dockerized
A lightweight Docker setup for running LangFlow as a reproducible visual environment for building and testing LLM, agent, and RAG workflows.
Docker . Python . LangFlow . LangChain . LLM Workflows
Tool: Dockerized PyTorch CNN Classifier API
A Dockerized Flask API that serves a trained PyTorch CNN for image classification, turning a notebook-trained model into a local inference service.
Python . PyTorch . torchvision . Flask . Docker . Pillow . CNN
Tool: Flask Web Scraper API
A Flask-based scraping and ingestion API that collects web pages, PDFs, and Google Drive documents, cleans them into structured JSON, and stores the results for downstream search or AI workflows.
Python . Flask . BeautifulSoup . Requests . PyPDF2 . MongoDB . Google Drive API . Langflow . JSON . CSV
Research Notes
The projects here are not only outputs. They are traces of questions: how to evaluate models honestly, how to clean data without losing meaning, how to turn notebooks into systems, and how to make technical work readable without flattening it.