AI Data Science & Machine Learning Course

Data Science and Python for ML Workflows

Focus: Core data manipulation and analysis using industry-standard Python libraries, at working-developer pace

NumPy for Numerical Computing

• Arrays, vectorization, and broadcasting: writing efficient, non-loop-based code

Pandas for Data Manipulation

• DataFrames/Series, advanced indexing and filtering
• Cleaning, merging, joining, grouping, pivoting, and reshaping datasets
• Performance considerations for larger datasets

Exploratory Data Analysis (EDA)

• Statistical summarization and distribution analysis
• Visualization with Matplotlib/Seaborn for diagnostic, not just presentation, purposes

Data Pipelines

• Ingesting data from Comma-Separated Values (CSV) files, Structured Query Language (SQL) databases, and Application Programming Interfaces (APIs); structuring reusable, reproducible data pipelines
• Preparing datasets for downstream ML use (train/test splits, encoding, scaling)

Hands-on Lab: Refactor a messy, real-world dataset into a clean, reusable data pipeline ready for modeling

Foundational ML Techniques

Focus: Foundational ML modeling techniques and the standard ML workflow

ML Problem Framing

• Supervised vs. unsupervised; regression vs. classification; choosing the right approach

Regression and Classification

• Linear/polynomial regression, logistic regression, k-Nearest Neighbors (k-NN), decision trees
• Evaluation metrics: Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R-squared (R²) for regression; accuracy, precision, recall, F1 score, and Receiver Operating Characteristic/Area Under the Curve (ROC/AUC) for classification

Feature Engineering

• Encoding, scaling, handling categorical and high-cardinality data
• Feature selection and leakage pitfalls

Model Validation

• Cross-validation, bias-variance tradeoff, avoiding overfitting
• scikit-learn pipelines for reproducible workflows

Hands-on Lab: Build, evaluate, and iterate on a classification pipeline in scikit-learn

Advanced ML Techniques

Focus: Advanced ML modeling techniques, tuning, and interpretability

Ensemble Methods

• Random forests and gradient boosting frameworks such as Extreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LightGBM); bagging vs. boosting

Unsupervised Learning

• Clustering (k-means, hierarchical); dimensionality reduction using Principal Component Analysis (PCA)

Hyperparameter Optimization

• Grid search, random search, and validation strategy at scale

Model Interpretability

• Feature importance and SHapley Additive exPlanations (SHAP) for explainability

From ML to Neural Networks

• Where classical ML breaks down (unstructured data, high-dimensional inputs)
• Conceptual bridge into neural network architectures

Hands-on Lab: Build and tune an ensemble model; compare against Day 2 baselines and interpret results

Deep Learning, Neural Networks and Architectures

Focus: Neural networks and deep learning architectures

Neural Network Mechanics

• Layers, weights, activation functions, forward/backpropagation, loss functions, optimizers
• Building and training networks in TensorFlow/Keras or PyTorch

Convolutional Neural Networks (CNNs)

• Convolution/pooling operations; image classification workflows

Sequence Architectures

• Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) networks for sequential and time-series data
• Where sequence models hit their limits, motivating the shift toward attention mechanisms

The Transformer Architecture

• Self-attention, positional encoding, and why Transformers displaced RNNs for language tasks
• Conceptual on-ramp into how LLMs are built on this architecture

Training in Practice

• Regularization, batch normalization, Graphics Processing Unit (GPU) acceleration, and training-time tradeoffs

Hands-on Lab: Train and evaluate a neural network on an image or text classification task

GenAI, Large Language Models (LLMs), Prompt Engineering, and Retrieval-Augmented Generation (RAG)

Focus: Practical application of LLMs, prompt engineering, and RAG

How LLMs Work in Practice

• Tokens, embeddings, and context windows: the mental model needed to build with LLMs
• Landscape of model providers and API access patterns

Prompt Engineering

• Prompt structure, system/user roles, few-shot examples
• Chain-of-thought, structured output, and iterative prompt evaluation

Building with LLM APIs

• Programmatic API calls, parameter tuning (temperature, max tokens), handling structured responses

Retrieval-Augmented Generation (RAG)

• Embeddings and vector databases (Chroma/FAISS/Pinecone)
• Designing a RAG pipeline: chunking strategy, retrieval, re-ranking, generation
• Evaluating RAG output quality and reducing hallucination

Production Considerations

• Cost/latency tradeoffs, data privacy, and where RAG fits into existing engineering systems

Hands-on Lab: Build a working RAG application over a custom document set, from ingestion through query response

Participants trace the full arc, from a raw Python data pipeline, through ML and Deep Learning models, to a deployed GenAI application.