Towards Intelligence

These are not prerequisites in the academic sense — they are the fastest path to becoming employable and understanding the landscape well enough to know where to go deep. The CS&E track gets you building real software with real tools (code, databases, version control, deployment). The Data & Intelligence track gets you working with data and AI systems end-to-end. Together, they give you enough working fluency to land a role and enough context to make the deeper sections below actually make sense.

Mathematics is not a separate track to “finish” — it is a foundation you keep revisiting. Probability and Statistics are essential for machine learning and data analysis. Linear Algebra is the language of neural networks and dimensionality reduction. Undergraduate Mathematics (calculus, real analysis) gives you the rigour to read research papers without hand-waving. The order here is roughly progressive: school-level refreshers first, then university-level depth, then the specific branches that matter most for AI and data work.

This discipline is split into Science and Engineering because they answer fundamentally different questions. Science asks: how does computation work? What makes an algorithm correct? Why does a distributed system fail in the ways it does? These are questions of understanding — and skipping them means you end up building things you cannot debug, optimise, or reason about when they break. Engineering asks: how do I build and run real systems? Knowing how a B-tree works is Science; designing a database schema that survives production traffic is Engineering. You need both, and in that order.

Within Science, the split is between Fundamentals and Systems. Fundamentals covers the abstract core — computation theory, algorithms, language design — the concepts that do not change regardless of what technology stack you use. Systems takes those ideas and grounds them in real infrastructure: how databases manage transactions, how operating systems schedule processes, how networks route packets. Fundamentals gives you the “what”; Systems gives you the “where it actually runs.”

Within Engineering, the split is between Building and Operating. Building is about construction — writing software, setting up infrastructure, designing cloud architectures, establishing CI/CD pipelines. But building a system is only half the job. Operating is about keeping it alive: ensuring reliability through SRE practices, catching performance regressions before users do, hardening security posture, and maintaining observability so that when something does go wrong at 3 AM, you can actually find out why. A system that cannot be operated is a system that cannot be trusted.

The same Science/Engineering split applies here, and for the same reason. Knowing the mathematics behind gradient descent (Science) and knowing how to debug a training run that is not converging (Engineering) are two entirely different skills — and you need both to be effective. The roadmap deliberately mirrors the two tracks topic-for-topic across the same five areas: Data Analysis, Machine Learning, Deep Learning, NLP, and Transformers/LLMs. The Science track uses canonical textbooks and Stanford lecture series to build theoretical depth — understanding why algorithms work, what their guarantees are, and where they break down. The Engineering track uses implementation-focused books and code repositories to build practical fluency — understanding how to actually train, evaluate, ship, and iterate on real models.

The Engineering track is further split into Building and Operating, just like in CS&E. Building covers Software Engineering (writing the model code, building the application, choosing the right framework) and Infrastructure Engineering (designing the ML system architecture, data pipelines, feature stores, and serving infrastructure). Operating covers the production lifecycle — monitoring model drift, managing retraining pipelines, ensuring reproducibility, and keeping ML systems reliable at scale. Building gets a model into production; Operating keeps it there.

The deliberate mirroring between Science and Engineering is the point. Theory without practice is academic — you understand the math but cannot ship anything. Practice without theory is fragile — you can follow a tutorial, but the moment something behaves unexpectedly, you have no mental model to fall back on. Both tracks together produce someone who can reason about and build intelligent systems.