Artificial Intelligence

You can manipulate tensors, compute matrix gradients, and evaluate probability distributions required for training models from scratch.

1. Linear algebra: vectors, matrices and tensors3 hrs
   • np.dot()
   • tensor shape broadcasting
   • eigenvalues and eigenvectors
   • singular value decomposition (SVD)
2. Calculus and optimization basics for AI3 hrs
   • partial derivatives
   • chain rule for backpropagation
   • gradient descent
   • scipy.optimize
3. Probability, statistics and Bayes' theorem3 hrs
   • probability density functions
   • maximum likelihood estimation (MLE)
   • Bayesian inference
   • scipy.stats
4. Propositional and predicate logic3 hrs
   • truth tables
   • first-order logic resolution
   • symbolic reasoning
   • satisfiability solvers
5. Discrete mathematics and graph theory for AI4 hrs
   • adjacency matrices
   • shortest path algorithms
   • combinatorics
   • networkx

Hands-on project: Implement a custom multi-layer linear algebra and gradient calculation engine in pure NumPy without high-level ML libraries.

You can formulate complex decision-making problems as state-space searches and implement optimal search algorithms.

1. What is AI: history, narrow vs general AI and the Turing test3 hrs
   • symbolic AI vs statistical AI
   • expert systems architecture
   • Turing test critique
   • AI ethics baseline
2. Intelligent agents and their environments3 hrs
   • reflex agents
   • utility-based agents
   • partial observability
   • rationality metrics
3. Problem-solving framed as search3 hrs
   • state-space graphs
   • node expansion
   • frontier management
   • heuristic design
4. Uninformed search: BFS, DFS and uniform-cost search3 hrs
   • queue and stack data structures
   • time complexity
   • space complexity
   • optimality guarantees
5. Informed/heuristic search: A* and greedy best-first4 hrs
   • admissible heuristics
   • consistent heuristics
   • priority queues
   • optimality proofs
6. Adversarial search: minimax and alpha-beta pruning4 hrs
   • zero-sum games
   • evaluation functions
   • alpha-beta cutoffs
   • depth-limited search
7. Constraint satisfaction problems3 hrs
   • backtracking search
   • arc consistency (AC-3)
   • forward checking
   • variable ordering heuristics
8. Local search: hill climbing, simulated annealing and genetic algorithms3 hrs
   • state landscapes
   • temperature schedules
   • crossover and mutation operators
   • fitness functions
9. Real-world AI applications across industries2 hrs
   • supply chain optimization
   • automated scheduling
   • fraud detection pipelines
   • logistics routing

Hands-on project: Build an intelligent pathfinding solver using A* search and adversarial minimax with alpha-beta pruning for a game state.

You can set up local and cloud development environments with Python, NumPy, Pandas, PyTorch, and TensorFlow.

1. Python essentials for AI4 hrs
   • list comprehensions
   • generator expressions
   • decorators
   • type hinting
2. NumPy, Pandas and Matplotlib4 hrs
   • vectorized operations
   • DataFrame groupby and aggregation
   • missing data imputation
   • matplotlib subplots
3. Jupyter, Google Colab and GPU setup4 hrs
   • jupyter lab extensions
   • colab runtime configuration
   • CUDA device management
   • pip and poetry dependency locks
4. Introduction to Scikit-learn, TensorFlow and PyTorch4 hrs
   • Estimator API
   • tf.keras.Model
   • torch.nn.Module
   • tensor device placement

Hands-on project: Configure a GPU-accelerated Jupyter environment on Google Colab and build a high-performance vector manipulation utility module.

You can design semantic ontologies, rule-based expert systems, and probabilistic reasoning models under uncertainty.

1. Logic-based knowledge representation4 hrs
   • propositional logic engines
   • description logics
   • knowledge bases
   • theorem provers
2. Rule-based and expert systems4 hrs
   • forward chaining
   • backward chaining
   • conflict resolution strategies
   • inference engines
3. Semantic networks and ontologies4 hrs
   • RDF graphs
   • OWL ontologies
   • SPARQL queries
   • knowledge graphs
4. Reasoning under uncertainty4 hrs
   • conditional probability tables
   • Bayesian belief networks
   • Markov random fields
   • exact and approximate inference

Hands-on project: Build a diagnostic expert system using Bayesian networks and first-order logic rules for automated fault isolation.

You can train, evaluate, and tune supervised and unsupervised machine learning models using Scikit-learn and XGBoost.

1. Supervised learning: regression and classification6 hrs
   • linear regression
   • logistic regression
   • support vector machines
   • decision trees
2. Unsupervised learning: clustering and dimensionality reduction6 hrs
   • K-means clustering
   • DBSCAN
   • Principal Component Analysis (PCA)
   • t-SNE visualization
3. Model evaluation, validation and the bias-variance trade-off6 hrs
   • k-fold cross-validation
   • ROC-AUC curves
   • precision-recall tradeoffs
   • regularization (L1/L2)
4. Ensemble methods: bagging, boosting and XGBoost6 hrs
   • Random Forest
   • Gradient Boosting Machines
   • XGBoost hyperparameter tuning
   • stacking classifiers

Hands-on project: Build an end-to-end predictive maintenance pipeline using feature engineering, hyperparameter tuning, and gradient boosting trees.

You can design, train, and debug deep neural networks, Convolutional Neural Networks, and recurrent architectures in PyTorch.

1. Neural networks, activations and backpropagation6 hrs
   • multi-layer perceptrons
   • ReLU and Sigmoid activations
   • forward pass mathematics
   • torch.autograd
2. Optimizers, regularization and batch normalization6 hrs
   • AdamW optimizer
   • dropout layers
   • batch normalization
   • learning rate schedulers
3. Convolutional Neural Networks (CNNs) for vision6 hrs
   • convolutional filters
   • max pooling
   • ResNet architecture blocks
   • transfer learning with torchvision
4. Recurrent networks: RNNs, LSTMs and GRUs6 hrs
   • hidden state dynamics
   • vanishing gradients
   • LSTM cell gates
   • sequence-to-sequence mapping

Hands-on project: Build an image classification and sequence modeling pipeline using custom PyTorch CNNs and Lstm networks with mixed-precision training.

You can build text classification, tokenization pipelines, and computer vision systems using OpenCV and spaCy.

1. Text preprocessing and tokenization3 hrs
   • regex cleaning
   • lemmatization
   • Byte-Pair Encoding (BPE)
   • spaCy pipelines
2. Word embeddings: Word2Vec and GloVe3 hrs
   • CBOW architecture
   • skip-gram model
   • cosine similarity
   • gensim library
3. Language models and the attention mechanism3 hrs
   • n-gram models
   • perplexity scoring
   • additive attention
   • alignment matrices
4. Sentiment analysis, named-entity recognition and machine translation3 hrs
   • token classification
   • sequence labeling
   • BLEU score evaluation
   • NLTK corpus tools
5. Image processing fundamentals with OpenCV3 hrs
   • color space conversions
   • thresholding
   • edge detection (Canny)
   • morphological operations
6. Object detection and recognition3 hrs
   • bounding boxes
   • Intersection over Union (IoU)
   • YOLO architecture basics
   • non-max suppression
7. Image segmentation3 hrs
   • semantic segmentation
   • instance segmentation
   • U-Net architecture
   • mask generation
8. Face detection and recognition3 hrs
   • Haar cascades
   • HOG features
   • facial landmark detection
   • embedding distance matching

Hands-on project: Create a real-time object detection and facial recognition system combined with an NLP sentiment analysis processing pipeline.

You can implement Transformer architectures, engineer prompts, build RAG pipelines, and fine-tune open-weights models.

1. The Transformer architecture and self-attention5 hrs
   • scaled dot-product attention
   • multi-head attention
   • positional encodings
   • encoder-decoder stacks
2. BERT, GPT and the modern LLM landscape5 hrs
   • masked language modeling
   • autoregressive decoding
   • decoder-only architectures
   • open-weights vs proprietary models
3. Prompt engineering fundamentals4 hrs
   • few-shot prompting
   • chain-of-thought reasoning
   • system prompts
   • structured JSON outputs
4. Fine-tuning and Retrieval-Augmented Generation (RAG)5 hrs
   • LoRA and QLoRA adaptation
   • vector embedding stores
   • hybrid search strategies
   • reranking retrieved chunks
5. Generative models: GANs, VAEs and diffusion5 hrs
   • discriminator-generator loops
   • latent space sampling
   • denoising diffusion probabilistic models
   • Stable Diffusion pipelines

Hands-on project: Build an enterprise Retrieval-Augmented Generation (RAG) system using Hugging Face Transformers, ChromaDB, and LangChain.

You can formulate Markov Decision Processes, train Q-learning agents, and orchestrate multi-agent LLM workflows.

1. Markov Decision Processes (MDPs)4 hrs
   • states and actions
   • transition probabilities
   • reward functions
   • discount factor gamma
2. Value iteration and Q-learning4 hrs
   • Bellman equation
   • temporal difference learning
   • epsilon-greedy exploration
   • Q-table updates
3. Policy gradient methods4 hrs
   • REINFORCE algorithm
   • policy networks
   • advantage estimation
   • reward shaping
4. Deep reinforcement learning (DQN and beyond)4 hrs
   • experience replay buffers
   • target networks
   • Deep Q-Networks (DQN)
   • PPO basics
5. Autonomous agents and decision-making4 hrs
   • ReAct framework
   • planning loops
   • reflection and critique
   • memory management
6. Multi-agent systems and coordination4 hrs
   • agent communication protocols
   • collaborative task breakdown
   • hierarchical multi-agent topologies
   • conflict resolution
7. LLM-powered agents and tool use4 hrs
   • function calling APIs
   • JSON schema validation
   • external API integration
   • sandbox execution
8. Planning, memory and orchestration with frameworks like LangChain4 hrs
   • vector memory stores
   • agent runtimes
   • state graph execution
   • custom tool wrappers

Hands-on project: Build an autonomous multi-agent coding assistant using LangChain, tool calling, and policy gradient reinforcement learning.

You can audit AI models for bias, implement explainable AI, containerize models, and deploy production endpoints.

1. Bias, fairness and transparency in AI3 hrs
   • disparate impact metrics
   • demographic parity
   • mitigating algorithmic bias
   • fairness constraints
2. Explainable AI (XAI)3 hrs
   • SHAP values
   • LIME feature attribution
   • saliency maps
   • interpretable surrogate models
3. AI safety and alignment2 hrs
   • constitutional AI
   • red teaming LLM prompts
   • guardrails against hallucination
   • content filtering
4. Privacy, regulation and AI governance2 hrs
   • differential privacy
   • federated learning
   • EU AI Act compliance
   • data anonymization
5. Saving, serializing and versioning models2 hrs
   • pickle and joblib serialization
   • ONNX runtime export
   • model artifact versioning
   • DVC data version control
6. Serving models with Flask and FastAPI2 hrs
   • Pydantic request validation
   • async model inference endpoints
   • batch prediction handlers
   • uvicorn ASGI server tuning
7. Containerization with Docker2 hrs
   • multi-stage Dockerfiles
   • CUDA base images
   • dependency caching
   • container security scanning

Hands-on project: Audit a machine learning model for disparate impact bias, generate SHAP explanations, containerize the FastAPI service with Docker, and configure MLflow model registry tracking.

You can serve a model behind an API and detect when it has started to degrade.

1. From notebook to service90 min
   • Model serialisation
   • FastAPI inference endpoint
   • Batch vs real-time
2. Versioning models and data90 min
   • Model registry
   • Reproducible training runs
   • Rollback to a prior model
3. Monitoring and drift detection90 min
   • Input and prediction drift
   • Reference windows
   • Alert thresholds
4. Evaluation that keeps working90 min
   • Offline vs online metrics
   • Shadow deployment
   • Human review loop

Hands-on project: Package a trained model behind a FastAPI endpoint in a container, add latency and prediction-distribution metrics to Prometheus, define a drift alert against a reference window, and write the runbook for what happens when it fires.

You can attack your own model and state honestly what it is and is not safe for.

1. Adversarial inputs and robustness90 min
   • Perturbation attacks
   • Robustness evaluation
   • Input sanitisation
2. Attacks specific to LLM systems90 min
   • Prompt injection
   • Indirect injection via retrieved text
   • Data exfiltration
3. Privacy and training data90 min
   • Memorisation and extraction
   • PII handling
   • Differential privacy basics
4. Documenting limits honestly90 min
   • Model cards
   • Bias and fairness measurement
   • Stating out-of-scope use

Hands-on project: Run an adversarial and prompt-injection test suite against a supplied model and a RAG endpoint, document every successful attack with its input, add input validation and output filtering, and publish a model card naming the residual limitations.

You can architect, deploy, and present an end-to-end production AI system while navigating technical hiring loops.

1. End-to-end AI capstone, deployed live4 hrs
   • system architecture design
   • cloud deployment on AWS/GCP
   • load testing inference endpoints
   • monitoring logs and metrics
2. Reading and implementing research papers3 hrs
   • arXiv paper dissection
   • translating math notation to PyTorch
   • reproducing baseline results
   • ablation study design
3. Kaggle competitions and leaderboard practice3 hrs
   • exploratory data analysis strategies
   • feature engineering pipelines
   • ensemble submission generation
   • Kaggle kernel optimization
4. Portfolio, GitHub and Kaggle profile building3 hrs
   • README documentation standards
   • clean repository structure
   • Kaggle Grandmaster portfolio tips
   • personal branding for AI engineers
5. Mock interviews, DSA refresh and HR rounds3 hrs
   • system design for AI applications
   • machine learning interview questions
   • behavioral interview preparation
   • salary negotiation tactics

Hands-on project: Design, build, and deploy a production-grade multi-agent RAG system to a live cloud cluster, pass technical AI mock interviews, and finalize your GitHub portfolio.