All Roadmaps

AI / ML Engineer

From mathematics and Python to training production ML models, deploying LLMs, and building end-to-end AI systems.

Mathematics for ML
Internal

Strong mathematical foundations are the difference between using ML tools and truly understanding them.

Read More
Linear Algebra

Vectors, matrices, eigenvalues, SVD, tensors — the language of neural networks.

Open
Calculus & Gradients

Derivatives, partial derivatives, chain rule, gradient descent, Jacobians.

Probability & Stats

Distributions, Bayes theorem, MLE, hypothesis testing, correlation vs causation.

Python for Data Science

Python is the lingua franca of AI/ML. Master the scientific computing stack.

NumPy

N-dimensional arrays, broadcasting, vectorisation, linear algebra ops, random number generation.

Open
Pandas

DataFrames, Series, groupby, merge, pivot, missing values, time series, reading CSV/JSON/Parquet.

Open
Matplotlib & Seaborn

Data visualisation — scatter, line, histogram, heatmap, pair plot, subplots, customisation.

Jupyter Notebooks

Interactive development, markdown cells, magic commands, Jupyter Lab, Google Colab.

Classical Machine Learning
Internal

Before deep learning — understand the algorithms that still power much of ML in production.

Read More
Scikit-learn

Fit, predict, pipelines, cross-validation, hyperparameter tuning, preprocessing.

Open
Supervised Algorithms

Linear/logistic regression, decision trees, random forests, gradient boosting (XGBoost).

Model Evaluation

Train/val/test split, k-fold CV, accuracy, precision/recall/F1, AUC-ROC, confusion matrix.

Feature Engineering

Encoding, scaling, imputation, feature selection, dimensionality reduction (PCA, t-SNE).

Deep Learning
Internal

Neural networks, backpropagation, and the architectures that power modern AI.

Read More
PyTorch

Tensors, autograd, nn.Module, DataLoader, training loop, GPU acceleration, torchvision.

Open
CNNs & Transfer Learning

Conv layers, pooling, ResNet/EfficientNet, torchvision models, fine-tuning pre-trained networks.

Transformers & Attention

Self-attention, multi-head attention, positional encoding, BERT/GPT architecture.

Optional Keras / TensorFlow

Sequential API, Functional API, model.fit, callbacks, TensorBoard, TFLite.

Open
Natural Language Processing (NLP)

Process, understand, and generate text — from tokenisation to fine-tuning large language models.

Hugging Face Transformers

Pre-trained models, tokenisers, pipelines, AutoModel, fine-tuning with Trainer API.

Open
Fine-tuning LLMs

PEFT (LoRA, QLoRA), Instruction tuning, RLHF basics, unsloth, LLaMA Factory.

RAG (Retrieval-Augmented Generation)

Vector embeddings, FAISS/Qdrant/Pinecone, chunking strategies, re-ranking, LangChain.

LangChain & Agents

Chains, tools, memory, agents (ReAct, function calling), LangGraph for multi-agent workflows.

Open
MLOps & Production ML

Train once, serve forever — versioning, monitoring, CI/CD for machine learning.

Experiment Tracking

MLflow, Weights & Biases (wandb) — log metrics, hyperparams, artefacts, compare runs.

Open
Model Serving

FastAPI + uvicorn, TorchServe, Triton Inference Server, vLLM for LLM serving, ONNX export.

Optional Data Pipelines

Apache Airflow, Prefect, Metaflow — DAGs, feature stores (Feast), data versioning (DVC).

Optional Model Monitoring

Data drift, concept drift, Evidently AI, Arize, performance degradation alerts.

Computer Vision
Optional

Object detection, image segmentation, and generation — the visual intelligence stack.

Optional Object Detection

YOLO v8, Faster R-CNN, DETR — bounding boxes, mAP metric, training custom datasets.

Optional Diffusion Models

Stable Diffusion, SDXL, ControlNet, LoRA fine-tuning, Diffusers library, image generation.

Optional OpenCV

Image preprocessing, edge detection, colour spaces, homography, camera calibration.

Open