AI Engineering

Modul 1: SQL for Data Science and Feature Engineering

• Goal: Build expert-level querying skills for data analysis, cleaning, and feature engineering
• SQL Foundations: SELECT, WHERE, GROUP BY, HAVING, ORDER BY, LIMIT, aliases
• JOINs: INNER, LEFT, RIGHT, FULL, SELF JOINs
• Subqueries, nested queries, scalar vs correlated subqueries
• Window functions: ROW_NUMBER, RANK, NTILE, LAG, LEAD, running totals
• CTEs (Common Table Expressions) and recursive queries
• Aggregations: COUNT, SUM, AVG, MIN, MAX, GROUPING SETS, ROLLUP, CUBE
• Filtering + cleaning: handling NULLs, deduplication, normalization in SQL
• A/B testing analysis: lift metrics, statistical significance from raw logs
• Building features in SQL: time-based features, event funnels, retention cohorts
• Query optimization basics (EXPLAIN plans)
• Mini Projects: Analyze marketing funnel events
• User churn prediction feature engineering
• Tools: PostgreSQL, SQLite, BigQuery, Mode/Redash

Modul 2: Python Programming Essentials for AI

• Goal: Write modular, maintainable, and production-ready Python code
• Python syntax & semantics: variables, loops, functions, scopes
• Data types: lists, tuples, sets, dicts, mutability, iteration
• Functional Python: map, filter, lambda, list comprehensions
• File I/O (CSV, JSON, YAML, pickle), error handling, context managers
• OOP: class design, inheritance, special methods (init, str)
• Virtual environments (venv, pipenv, poetry), package management
• Code structuring: modules, packages, docstrings, type hints
• Logging, debugging, assertions
• Unit testing with pytest, test coverage
• CLI tools with argparse
• Intro to working with APIs (e.g., requests)
• Mini Project: A CLI-based ETL pipeline pulling data from a public API and saving to CSV/DB
  
  

Modul 3: Data Structures & Algorithms for ML Engineers

• Goal: Enhance problem-solving and implementation skills essential for system design and model performance
• Sorting: quicksort, mergesort, bubble, heap sort
• Searching: binary search, hash tables
• Data Structures: stacks, queues, heaps, linked lists
• Advanced: tries, graphs, trees (binary, BST), hash maps, sets
• Graph algorithms: DFS, BFS, Dijkstra
• Algorithm paradigms: greedy, divide & conquer, DP
• Understanding complexity: time/space trade-offs (Big O notation)
• Practice: Daily LeetCode challenges
• Interview-style problem solving (Python-based)
• Special Focus: LRU cache
• Real-time recommender system structure
• Approximate nearest neighbor for vector search
  
  

Modul 4: Data Analysis, Wrangling, and Visualization

• Goal: Prepare data for modeling through EDA and advanced wrangling techniques
• NumPy: broadcasting, matrix multiplication, indexing, random sampling
• Pandas: merge, groupby, pivot, apply, lambda ops, categorical encoding
• Handling missing data, duplicates, inconsistent types
• Datetime handling, resampling, time windowed aggregations
• Feature extraction & transformation: binning, normalization, scaling
• Visualization with Matplotlib / Seaborn / Plotly:
• Line/Bar/Box plots
• Correlation heatmaps, pairplots
• Interactive dashboards (Plotly Dash basics)
• Project: EDA on a real-world multi-source dataset
  Tools: Pandas, NumPy, Matplotlib, Seaborn, Pandas-Profiling, Sweetviz

Modul 5: Math, Statistics & Probability for Machine Learning

• Goal: Build mathematical foundations to understand and debug AI models
• Probability: Discrete/continuous RVs, conditional probability, independence
• Bayes Theorem, Naive Bayes classification
• Descriptive Stats: variance, std, IQR, skewness, kurtosis
• Inferential Stats:
• Confidence intervals
• Hypothesis testing: t-tests, z-tests, chi-square, ANOVA
• Linear Algebra:
• Vectors, matrices, transpose, dot product
• Projections, eigenvectors/values
• Singular Value Decomposition (SVD)
• Calculus:
• Derivatives, partial derivatives, gradients
• Chain rule, Taylor expansion
• Optimization:
• Gradient descent, learning rate schedules
• Convex vs non-convex functions
• Visual Demos: Logistic regression gradient visualization
• PCA step-by-step decomposition
  Tools: SymPy, Matplotlib, JAX (optional)

Modul 6: Machine Learning with Scikit-learn

• Goal: Master end-to-end ML systems with theory, code, and explainability
• End-to-end pipeline: cleaning, encoding, scaling, modeling
• Regression: Linear, Lasso, Ridge, ElasticNet, Poisson, Polynomial
• Classification: Logistic, KNN, SVM, Decision Trees, Random Forest, GBDT (XGBoost)
• Unsupervised: KMeans, DBSCAN, hierarchical clustering
• Dimensionality reduction: PCA, UMAP, t-SNE
• Evaluation metrics: F1, ROC-AUC, PR Curve, log-loss
• Feature engineering: interaction terms, time-based features
• Explainability:
• Permutation importance
• SHAP & LIME
• Model selection: grid/random search, nested CV
• Model persistence: joblib, pickle
• Projects:Tabular classification: Churn, Titanic 
• Regression: House prices
• NLP: TF-IDF + Logistic for sentiment
  Tools: Scikit-learn, XGBoost, SHAP

Modul 7: Deep Learning & TensorFlow (Foundations)

• Goal: Build deep learning models with strong theoretical grounding
• Tensors, computation graphs, sessions (TF2.x)
• Forward/backward propagation manually + Keras
• Neural network architecture: layers, activations
• Loss functions: MSE, binary cross-entropy, categorical CE
• Optimizers: SGD, Adam, learning rate decay
• Regularization: Dropout, L2, early stopping
• Batch normalization, weight initialization
• TensorBoard logging, callbacks
• Model evaluation: confusion matrix, ROC in deep learning
• Saving/loading: SavedModel, checkpoints

• Project: Build a 3-layer NN to classify MNIST/Fashion-MNIST

• Tools: TensorFlow, Keras, TensorBoard
  
  

Modul 8: Computer Vision (CV)

• Goal: Build and fine-tune CNNs for image classification & detection
• Convolutions: filters, padding, stride, receptive fields
• Pooling: max, average, global
• CNN Architectures: LeNet, VGG, ResNet (skip connections)
• Transfer Learning: MobileNet, EfficientNet
• Fine-tuning vs Feature extraction
• Data Augmentation: flips, rotations, random crop
• Image normalization, pre-trained weights
• Intro to object detection: YOLO overview
• Working with OpenCV for preprocessing
• Live video processing (optional)
• Project: Waste/image classifier using EfficientNet or ResNet
  Tools: TensorFlow, OpenCV, Albumentations

Modul 9: NLP (Traditional & Neural)

• Goal: Build NLP systems using both classical and deep methods
• Text normalization, tokenization, stopword removal
• N-grams, TF-IDF, cosine similarity
• Word embeddings: Word2Vec, GloVe
• Classical models: Naive Bayes, Logistic, SVM
• RNNs, GRUs, LSTMs for sequence modeling
• Sequence padding, masking
• Attention mechanism basics
• Evaluation: precision, recall, BLEU, ROUGE
• Named Entity Recognition, POS tagging
• Project: Text classifier (news/spam/sentiment)
• Tools: spaCy, Gensim, TensorFlow, NLTK
  
  

Modul 10: Transformers & Hugging Face

• Goal: Master transfer learning with LLMs and modern NLP tools
• Transformer architecture: self-attention, multi-head attention, encoder/decoder
• Pre-trained models: BERT, DistilBERT, RoBERTa, GPT-2
• Tokenizers: fast vs slow, Byte-Pair Encoding
• Fine-tuning on custom classification, QA, summarization
• Push & load from Hugging Face Hub
• Dataset preprocessing with datasets
• Prompt tuning and low-rank adaptation (LoRA)
• Projects:
• Summarization using BART
• QA bot using DistilBERT
• Tools: Hugging Face Transformers, Datasets, Accelerate

Modul 11: Model Deployment & MLOps

• Goal: Deploy, monitor, and manage ML models in production
• Serialization: joblib, pickle, SavedModel
• Building REST APIs with FastAPI
• Creating interactive UIs with Streamlit or Gradio
• Dockerizing models and APIs
• Tracking models with MLflow
• CI/CD for model deployments (GitHub Actions)
• Monitoring models in prod: drift detection, logging
• Scheduling retraining jobs (with Airflow/Prefect intro)
• Project: Full-stack deployment of any prior model with a UI + API + Docker
• Tools: FastAPI, Streamlit, Docker, MLflow