Track 01 — The Neural Wing
Artificial Intelligence
From your first Python notebook to a model running behind a real interface. You will learn the maths you actually need, build classical and deep learning models, then put a large language model to work on a problem that matters — and ship it.
( What you'll learn )
Python for data
The language properly, then NumPy and pandas — loading, cleaning and reshaping messy real-world data without fighting it.
The maths you actually need
Linear algebra, probability and statistics at the depth a working practitioner uses. Enough to reason about models, not a pure maths degree.
Classical machine learning
Regression, decision trees, ensembles and clustering — plus the harder skill of evaluating a model honestly instead of celebrating a leaked score.
Feature engineering
Where most of the real gains hide. Encoding, scaling, leakage, imbalance and how to build a pipeline you can reproduce.
Deep learning
Neural networks from the ground up, then convolutional nets for vision. Training loops, overfitting, regularisation and why your loss is not going down.
NLP & transformers
Tokenisation, embeddings, attention and the transformer architecture — then fine-tuning a pretrained model on your own data.
LLM applications
Prompt design, retrieval-augmented generation, tool use and agents. How to build something on top of an LLM that is actually reliable.
Evaluation & testing
How to know whether your system is good. Benchmarks, human review, regression suites, and catching quality drops before users do.
MLOps essentials
Experiment tracking, model versioning, packaging behind an API, containers and monitoring in production.
Ethics, bias & safety
Where models cause harm and how to reduce it — dataset bias, privacy, explainability, and knowing when not to automate a decision.
Working with GPUs
Free and low-cost compute, notebooks, memory limits and the practical tricks for training on hardware you can afford.
Capstone project
One substantial problem taken end to end — data, model, evaluation, interface, deployment — presented and defended in front of the group.
( Curriculum — 26 weeks )
Foundations
Getting genuinely comfortable with Python and data before any model appears. Most people who struggle later struggled here first.
- Python syntax, functions, classes
- NumPy arrays and vectorised thinking
- pandas: load, clean, join, reshape
- Visualisation with matplotlib
- Linear algebra for ML
- Probability and statistics
- Jupyter, virtual environments, Git
- Your first end-to-end analysis
Machine learning
The whole classical toolkit, with a heavy emphasis on evaluating results honestly and building pipelines you can rerun.
- Supervised learning: regression, classification
- Trees, random forests, gradient boosting
- Unsupervised: clustering, dimensionality reduction
- Cross-validation and data leakage
- Metrics: precision, recall, ROC, calibration
- Imbalanced data strategies
- Hyperparameter search
- scikit-learn pipelines
Deep learning
Neural networks built by hand first so the framework stops being magic, then PyTorch for real work.
- Perceptrons, backpropagation, gradients
- PyTorch tensors and autograd
- Training loops and optimisers
- Convolutional networks for vision
- Transfer learning
- Data augmentation
- Overfitting, dropout, early stopping
- Experiment tracking
LLMs, deployment & capstone
Modern applied AI, then getting your work out of a notebook and in front of a user.
- Transformers and attention
- Fine-tuning pretrained models
- Embeddings and vector search
- Retrieval-augmented generation
- Tool use and agent patterns
- Serving a model behind FastAPI
- Docker and cloud deployment
- Capstone build and presentation
( Tools you'll master )
( Projects you'll build )
Churn prediction with an honest write-up
A full tabular ML project: cleaning, features, model comparison, and a report explaining what the numbers mean and where the model fails.
Image classifier on your own dataset
Collect and label your own images, train a CNN with transfer learning, and analyse the mistakes it makes rather than just the accuracy.
Document Q&A with retrieval
A RAG system over a real document set — chunking, embeddings, retrieval, and evaluation of whether the answers are actually grounded.
Deployed capstone
Your own problem, taken all the way: model behind an API, a simple interface, containerised and live on a URL you can put on a CV.
( Where this leads )
( Is this you? )
A good fit if
- You can commit real practice hours outside class — this track rewards repetition
- You are a graduate, developer or analyst wanting to move into AI work
- You are comfortable being confused for a week before something clicks
- You want to understand models, not just call an API and hope
Probably not yet if
- You have never written code at all — start with Full Stack or SEO first
- You want a certificate quickly with minimal effort
- You are hoping to skip the maths entirely; we keep it practical but it is not optional
- You cannot attend or catch up on sessions for weeks at a time
Built, not watched.
Every module in this track ends in something that runs. You will finish with four projects you can open in front of an interviewer and explain line by line — including the parts that did not work and what you did about them. That conversation is what gets people hired.
A note on honesty: we do not promise placements. What we promise is that you will leave able to build and defend real AI work, with mentors who review it properly along the way.
( Reading on this track )
Next batch enrolling
Enroll ↗