Data, MLOps & Deployment
The full arc of applied machine learning: asking a decision-shaped question and exploring data, engineering and selecting features, building pipelines and recommenders, then shipping, serving and monitoring a model that stays healthy as the world drifts. Covers the data-science, data-engineering, MLOps and time-series material an M.Tech AI programme expects, in the same house style: a concrete hook, a visualization you can play with, the math, then honest caveats.
Data science foundations
What data science really is, how to look at a dataset before modelling it, and how to reason from a sample to a population without fooling yourself.
- What data science actually isNot a toolbox but a loop: turn a decision into a question, data into evidence, and evidence into a choice someone can defend. 9 min · intro
- Exploratory data analysisPlots before models: distributions, outliers, and the summary statistics that lie about them. 10 min · intro
- Statistical inferenceSample means have their own distribution; the standard error and the confidence interval are how we say how wrong a sample might be. 11 min · core
Features & recommenders
Turning raw columns into signal: generating features, selecting the ones that earn their place, and making recommendations from a ratings matrix.
- Feature generationA model can only see the coordinates you give it. Good features move the signal into a space the model can use. 11 min · core
- Feature selectionMore features are not better. Filters rank them cheaply; wrappers search for the combination that actually helps. 11 min · core
- Recommendation systemsFrom a sparse ratings matrix: neighbourhood similarity, matrix factorisation, and the cold-start problem no algorithm escapes. 12 min · core
Data engineering
Pipelines as DAGs, data quality and lineage, and handling unstructured text and images at scale.
- Data pipelinesA pipeline is a DAG of idempotent steps with explicit dependencies — and the wall-clock time is its longest path, not the sum of its parts. 11 min · core
- Data quality and lineageTrust is a measurement, not a feeling: define quality dimensions, score them per table, and trace every downstream table a broken source can reach. 11 min · core
- Unstructured dataText, images and audio do not arrive as columns. Encoding them into vectors is the bridge between raw content and every model downstream. 11 min · core
MLOps
The ML lifecycle, deployment strategies that limit blast radius, and drift monitoring that tells you when a shipped model has stopped being right.
- The ML lifecycleFraming, data, modelling, deployment, monitoring — a loop with memory, where shipping is the midpoint rather than the finish line. 11 min · core
- Deployment and servingCanary, blue-green and rolling releases; batch versus online serving; and the rollback that turns a bad version into a non-event. 12 min · core
- Monitoring and driftData drift, concept drift and the PSI alarm — how you find out a model has gone stale before your users tell you. 12 min · advanced
Time series
Autocorrelation, stationarity and forecasting — why the order of observations carries information that a shuffled train-test split destroys.