Skills and Responsibilities
The rent model prices flat C at ₹26 thousand, and the owner phones to ask why. Answering well needs maths (what the model computed), data knowledge (what it learned from) and plain speaking. An ML engineer's skills are wide because the job touches everything between raw data and a customer's question.
The rule
Maths to understand, code to build, data to feed, judgement to ship.
Each part of the rule is a family of skills:
- Maths to understand. Linear algebra holds the data: each flat is a list of numbers, the five flats form a table. Calculus tells training which way to nudge each weight. Probability and statistics say how sure you can be. These are the next four sections of this course.
- Code to build. Python, tests, version control, and turning notebooks into scripts that run unattended.
- Data to feed. Querying databases, cleaning messy fields, building pipelines, and noticing when data changes.
- Judgement to ship. Choosing metrics, spotting a model that only looks good, explaining results to people who do not code, and knowing when not to use ML at all.
Nobody starts strong in all four. Most ML engineers arrive from software, from data analysis or from a maths-heavy degree, and spend their first years filling in the other families. Knowing which family you are weakest in tells you what to study next.
The responsibilities follow the lifecycle from the first lesson: frame the problem, gather and clean data, train, evaluate, deploy, monitor, and explain.
The shortcut
For any responsibility, name the skill it leans on, then name what breaks without it. If you can do both, you understand the duty.
The duty
Explain a single prediction: why ₹26 thousand?
Without the maths step the engineer cannot say what drove the price. Without the judgement step the answer is correct but useless to the owner.
Drills
Answer fast — trust the rule.
The owner of flat C asks why the model suggested ₹26 thousand. Which skill matters most here?
Which area of maths describes each flat as a list of numbers and the five flats as a table?
Which area of maths tells training which way to nudge each weight?
A pipeline silently drops listings with no metro distance, so far-away flats vanish from training. Whose problem is this?
Remember
- Maths to understand, code to build, data to feed, judgement to ship.
- Shortcut: for each duty, name its skill and what breaks without it.
- Trap: thinking the job is only training models; most failures start in data or judgement.