AI Engineer Master Program

Variables, data types, operators, input/output, strings, lists, tuples, sets and dictionaries.

if/elif/else, loops, break, continue and comprehensions.

Definitions, parameters, return values, scope, lambdas and higher-order concepts.

Classes, objects, constructors, methods, inheritance, encapsulation, polymorphism and abstraction.

Exceptions, files, JSON, modules, packages, decorators, generators, iterators, virtual environments and typing.

Logging, configuration, environment variables, dependency management, project structure and testing fundamentals.

Practical scalars, vectors, matrices, tensors and matrix operations.

Vector operations, dot products, matrices, multiplication, transpose, dimensions and eigenvalue concepts.

Functions, slope, derivatives, gradients, partial derivatives and optimization intuition.

Probability, conditional probability, independence, Bayes theorem and distributions.

Mean, median, mode, variance, standard deviation, percentiles, correlation, covariance, sampling and hypothesis concepts.

Arrays, dimensions, shape, indexing, slicing, broadcasting and vectorization.

Matrix operations, statistics, random numbers, normalization and numerical computation.

Series, DataFrames, CSV, JSON and Excel loading, selection and filtering.

Missing values, duplicates, invalid records, type conversion and outliers.

Grouping, aggregation, merging, joining, pivoting and reshaping.

Categorical and numerical variables, encoding, scaling and normalization.

Line, bar, scatter, histogram and box plots.

Identify trends, outliers, correlations, distributions and class imbalance.

SELECT, WHERE, ORDER BY, GROUP BY, HAVING and aggregate functions.

INNER, LEFT, RIGHT concepts and self joins.

Subqueries, CTEs, window functions, indexes, transactions and optimization concepts.

Design and query databases for production AI applications.

Features, labels, training, inference, models, parameters and hyperparameters.

Data -> preparation -> feature engineering -> training -> validation -> evaluation -> deployment -> monitoring.

Training, validation and test sets, leakage and cross-validation.

Regression, coefficients, errors, loss and prediction.

Classification, probabilities and decision boundaries.

Splitting, entropy and information-gain concepts and overfitting.

Ensemble learning, bagging and feature importance.

Distance, neighbors and classification.

Margins, hyperplanes and kernel concepts.

Gradient boosting plus XGBoost and LightGBM concepts.

K-Means, cluster selection and silhouette concepts.

Dendrogram and linkage concepts.

Density-based clustering and anomaly-detection concepts.

PCA, feature compression and visualization.

Accuracy, precision, recall, F1, confusion matrices, ROC and AUC concepts.

MAE, MSE, RMSE and R-squared.

False-positive and false-negative costs, revenue impact and operational impact.

Missing values, categorical encoding, normalization, standardization, interactions, date and text features.

Correlation, feature importance, dimensionality and leakage.

Scikit-learn preprocessing, training and prediction pipelines.

Grid search, random search, cross-validation and tuning strategy.

Regularization, cross-validation, feature reduction and early-stopping concepts.

Neurons, input, hidden and output layers, weights and biases.

ReLU, sigmoid, tanh and softmax.

How data flows through neural networks.

Conceptual error, gradients and weight updates.

Gradient descent, SGD, Adam and learning rate.

Regression, classification and cross-entropy losses.

Tensors, operations, datasets, DataLoader and models.

Layers, forward pass, training loop and validation loop.

CPU vs GPU, GPU tensors and training acceleration.

Pixels, channels, RGB, image dimensions and preprocessing.

Read, resize, crop, filter and transform images.

Convolution, filters, feature maps and pooling.

Build and evaluate an image classifier.

Adapt pretrained models instead of training from scratch.

Bounding boxes, detection, confidence and IoU concepts.

Text, tokens, vocabulary, preprocessing and normalization.

Bag of words, TF-IDF and n-grams.

Semantic similarity and dense representations.

Sentiment, spam and ticket classification.

Conceptual RNN, LSTM and GRU.

Attention, self-attention, query, key, value, multi-head attention and positional encoding.

Encoder-only, decoder-only and encoder-decoder architectures.

Tokenizers, pretrained models, pipelines and dataset concepts.

Classification fine-tuning, configuration, validation and inference.

Generative models, LLMs, foundation models, inference and context windows.

Authentication, requests, messages, structured output, streaming, retries and rate limits.

Zero-shot and few-shot prompting, instructions, constraints, context, templates and structured output.

System, retrieved, conversation, tool-output and memory context.

Vector representations, semantic meaning, similarity and cosine similarity.

FAISS, pgvector and managed vector-database concepts.

Combine keyword and vector search.

Document -> parse -> chunk -> embed -> vector store -> retrieve -> LLM -> grounded response.

PDF, DOCX, TXT, HTML and CSV support.

Fixed, recursive and semantic chunking concepts.

Top-K, metadata, filters, reranking, query rewriting and hybrid retrieval.

Retrieval and answer relevance, groundedness, faithfulness and citations.

Goals, tools, state, memory, actions and observations.

Search, calculator, database, APIs and knowledge-base tools.

Router, planner-executor, supervisor and human-approval patterns.

Models, prompts, tools, retrievers and agents.

Nodes, edges, state, conditional workflows, persistence and human-in-the-loop.

Prediction, classification, generation and RAG query endpoints.

Request and response schemas and validation.

async/await, parallel inference and streaming.

Client -> FastAPI -> model -> prediction.

Notebook -> package -> API -> container -> deployment -> monitoring.

Saving, loading, versions and artifacts.

Scheduled high-volume predictions vs real-time request prediction.

Code + data + model + infrastructure.

Models, datasets, parameters, metrics and artifacts.

Runs, experiments, artifacts and model registry.

DEVELOPMENT -> VALIDATED -> APPROVED -> PRODUCTION -> RETIRED lifecycle.

Dataset versions, reproducibility and lineage.

Managed fraud-model-v1, v2 and v3 style versions.

Code -> tests -> model evaluation -> build.

Approved model -> container -> staging -> validation -> production.

Block deployment below accuracy, fairness, security or required-evaluation thresholds.

Requests, predictions, latency, failures and model version.

Detect production inputs changing from training data.

Detect changing relationships between inputs and targets.

Alert when production model quality drops.

Tokens, latency, model errors, cost, RAG quality and safety signals.

Request ID, model and version, latency, status and errors.

Throughput, latency, error rate, prediction distribution and resource use.

Trace user -> API -> retriever -> model -> tool -> response.

Dataset and model bias, representation and fairness concepts.

Feature importance, SHAP concepts and explainable predictions.

Data minimization, consent, access controls and sensitive information.

Require human review for appropriate AI decisions.

Model, API, data and Generative AI security.

Adversarial input, model manipulation and poisoning concepts.

Prompt injection, malicious documents, data leakage, tool misuse and output validation.

Authentication, authorization, input validation, rate limits and audit logging.

Images, containers, Dockerfiles, ports, volumes and environment variables.

Containerize a Python application, model and FastAPI service.

Containerize FastAPI, RAG and PostgreSQL/vector-store components.

IAM, EC2, S3, RDS, networking, monitoring and secrets.

Cloud object storage for datasets, models and artifacts.

Deploy and operate an AI API.

Load balancer -> AI API instances -> model/service -> database -> monitoring.

Choose model, data and training/API strategy; analyze latency, scale, monitoring, cost and security.

Architecture and trade-off discussion for recommendation systems.

Streaming data, prediction, thresholds and monitoring.

Upload -> processing -> embedding -> retrieval -> generation.

Image -> preprocessing -> model -> prediction -> storage -> monitoring.

CPU, GPU, memory, storage and network cost.

Self-hosted vs managed APIs and smaller vs larger models.

Input/output tokens, caching and model routing.

Combine a traditional ML prediction service, a deep-learning image or NLP model, an LLM assistant, enterprise RAG, a tool-based agent, FastAPI, PostgreSQL, model versioning, Docker/AWS deployment, logs, metrics, authentication and authorization.