Data Warehousing & Business Intelligence - course datasets, assignment briefs and marking guides

Data Visualization course — lecture notebooks and materials

Two teaching demos: a RAG chatbot and a LangGraph agent, both on the Gemini API. Fully logged, no black boxes.

What a sentiment model costs you on the text it was not trained on: four corpora, every model evaluated on every one of them.

Same 1.9M NYC crashes, same question: r=0.05 between crashes, 0.55 across community districts, 0.71 across boroughs -- and equal-size contiguous redistricting alone spans 0.07 to 0.83. A measured a…

Your A/B test winner is about half as good as it looked. The winner's curse, peeking and power measured on 32,363 real randomised experiments -- with a live demo that re-splits the traffic in your …

Does a CrewAI crew write better SQL than one model call? 26 questions over the full Formula 1 record: the crew wins by one question for 4.8x the LLM calls -- and only if your harness forgives the f…

Uniform noise scores 0.925 point-adjusted F1 on the Numenta Anomaly Benchmark, beating three of five real detectors. An audit of the metric time-series anomaly detection reports, and a replacement …

An A2A agent network that reviews Python dependencies: four agents, one protocol, and a benchmark that measures whether the delegation was right

Multi-label emotion classification on GoEmotions: a 3.8 MB linear model at 96% of BERT, served from FastAPI and in-browser JS

Do you need a transformer to forecast? Five tiny models (none over 80k params) vs published transformer results on the ETTh1 long-horizon benchmark — with an in-browser demo.

MCP server that puts Socrata open data on tap for LLM clients — catalog search, schema inspection, and guarded SoQL queries, with a live in-browser demo

Movie recommender with hand-rolled implicit ALS: temporal leave-last-out evaluation, popularity-bias metrics, and a live fold-in demo on MovieLens 1M

Deep learning traffic-sign recognition with leakage-aware evaluation, confidence calibration and explicit abstention: PyTorch transfer learning, TorchScript + FastAPI serving, model card

How long does New York City take to fix things? Survival analysis of 1.77M 311 service requests: Kaplan-Meier for right-censored open requests, direct standardization for complaint mix

Multi-agent research brief generator: a LangGraph pipeline of planner, researcher, synthesizer, critic and writer agents that produces cited markdown briefs grounded in Wikipedia

RAG chatbot grounded in the official Python PEPs — section-aware chunking, local MiniLM embeddings, FAISS retrieval with ground-truth evaluation, and Gemini answers with parsed [PEP N] citations

🧠 Complete AI roadmap from zero to expert — free courses, videos, articles, and hands-on Kaggle/GitHub projects for every level

100 days of data science projects: Computer Vision, NLP & Machine Learning — Jupyter notebooks with full reports and datasets. New projects daily.

End-to-end ML pipeline forecasting hourly bike-share demand: EDA, leakage-guarded features, time-aware model selection, honest held-out evaluation

End-to-end data analysis: which countries decoupled economic growth from CO2 emissions? Reproducible pipeline on Our World in Data, with consumption-based accounting.

Self-hosted uptime monitoring service: async HTTP checks, SQLite storage, uptime/latency stats, and incident tracking behind a FastAPI API.

End-to-end customer churn prediction service: data validation, CV model selection, probability calibration, and an observable FastAPI inference API with Docker and CI.

Illuminating Islamic Knowledge Through AI — A conversational AI assistant that answers Islamic questions in Roman Urdu with verified citations from Quran, Hadith, and Tafseer

Intelligent Blood Classifier, Chat & Hospital Finder