AI Internship Prerequisites for PSG College of Technology Students: Math, Python and ML Basics

Before a short AI internship, a PSG College of Technology student needs four things: working comfort with vectors and matrices, basic probability and statistics, Python with NumPy and pandas, and the ability to train and evaluate one simple machine learning model. You do not need a full course in each. You need enough to read a small notebook, change it, and explain what the output means.

This guide gives you a self-assessment scorecard, a week-by-week plan, free practice exercises and a worked example. It is an independent guide for students. It is not published by, or connected with, the college, so the course-mapping section tells you how to match these topics to your own syllabus rather than assuming what your department teaches.

What “prerequisite” means for a two-week AI internship

A short internship is not the place to learn what a matrix is. Fourteen days is enough to complete a small guided task, such as cleaning a dataset, training a baseline classifier and writing up the result, but only if the basics are already in place. Students who arrive without them tend to spend the first week on setup and notation and the second week rushing.

The free two-week online internship from CareerSwami is one option for applying these basics. It offers a free offer letter and a free completion certificate after successful completion, and it runs across 26 tracks. Whichever programme you choose, the preparation below transfers.

The four pillars

• Linear algebra: data is stored as vectors and matrices, and most models are matrix operations.
• Probability and statistics: models output probabilities, and evaluation depends on understanding variance, distributions and sampling.
• Python: the working language of nearly every beginner AI task.
• Machine learning basics: the train, validate and test workflow, and the common failure modes.

Prerequisite scorecard: rate yourself honestly

Score each row from 0 to 2. Zero means you have not seen it. One means you recognise it but cannot do it unaided. Two means you can do it on a blank page or notebook without searching. A total of 28 or more out of 40 (20 rows) is a sensible readiness target; below 20, spend a few weeks on the plan further down before applying. These thresholds are a practical rule of thumb, not an official standard.

Pillar 1: Linear algebra

Focus on mechanics and shapes. Most beginner bugs in AI code are shape mismatches, so being fluent here pays off quickly.

What to cover

• Vectors as lists of features; addition and scalar multiplication.
• Dot product as a similarity measure and as the core of a linear model: prediction = w·x + b.
• Matrices as datasets (rows are samples, columns are features).
• Matrix multiplication and its shape rule: (m×n)(n×p) gives (m×p).
• Transpose, identity, inverse (concept only).
• Eigenvalues and eigenvectors at an intuition level, so that PCA and SVD are not mysterious.

Free practice

1. Do ten dot-product and matrix-multiplication problems by hand, then verify each with numpy.dot.
2. Build a 5×3 array of random numbers. Predict the shape of X.T @ X before you run it.
3. Write a function that computes cosine similarity between two vectors without using a library function.
4. Work through the free lecture and problem sets on linear algebra from MIT OpenCourseWare or the Khan Academy linear algebra course, choosing the first four weeks of topics only.

Pillar 2: Probability, statistics and a little calculus

What to cover

• Sample space, events, independence, conditional probability.
• Bayes’ theorem, which underlies Naive Bayes and the way classifier outputs should be read.
• Random variables, expectation and variance.
• Bernoulli, binomial, uniform and normal distributions.
• Sampling, bias and why a small sample misleads.
• Derivatives, partial derivatives and the gradient, enough to follow what gradient descent does.

Free practice

1. A disease affects 1 in 100 people. A test is 95% sensitive and 90% specific. Compute the probability that a positive person is actually ill. Most people guess far too high; the right answer is a little under 9%.
2. Simulate 1,000 coin flips in Python at three different bias values and plot the running proportion of heads.
3. Generate 10,000 samples from a normal distribution, plot the histogram, and check how many fall within one and two standard deviations.
4. Differentiate f(w) = (w − 3)² by hand, then find the minimum by following the negative gradient for ten steps in a loop.

Pillar 3: Python for AI

Aim for fluent, readable code rather than clever code. Internship reviewers read notebooks, and clear variable names and short functions matter more than tricks.

What to cover

• Core Python: types, control flow, functions, list and dictionary comprehensions, file handling, exceptions.
• Virtual environments and installing packages with pip.
• Jupyter or Google Colab, including restarting and running all cells to confirm a notebook reproduces.
• NumPy: arrays, indexing, slicing, broadcasting, vectorised operations.
• pandas: read_csv, describe, filtering, groupby, handling missing values.
• matplotlib or seaborn for basic charts.
• Git basics: init, commit, push to a public repository.

Free practice

1. Pick any public CSV dataset (the UCI Machine Learning Repository and Kaggle both host free ones) and answer five questions about it using only pandas.
2. Rewrite a loop-based calculation of column means as a single NumPy expression, and compare the time taken on a large array.
3. Write a function that takes a DataFrame and returns the percentage of missing values per column, sorted descending.
4. Push one finished notebook to GitHub with a README of five lines: the question, the data source, the method, the result and one limitation.

Pillar 4: Machine learning basics

What to cover

• Supervised versus unsupervised learning; regression versus classification.
• The workflow: define the target, split the data, build a baseline, train, evaluate, iterate.
• Linear and logistic regression, k-nearest neighbours, decision trees and k-means.
• Metrics: MAE and RMSE for regression; accuracy, precision, recall, F1 and the confusion matrix for classification.
• Overfitting, underfitting, cross-validation, and data leakage.
• Feature scaling and encoding categorical variables.

Free practice

1. Train a decision tree and a logistic regression on the same dataset with scikit-learn. Report both on the same test split and say which you would choose and why.
2. Deliberately cause overfitting by setting a tree’s depth very high. Plot training versus validation score as depth grows.
3. Create leakage on purpose by scaling before splitting, then fix it with a pipeline. Write two sentences on what changed.
4. Take an imbalanced dataset, show that always predicting the majority class gives high accuracy, and then report recall for the minority class.

Worked example (illustrative): one gradient descent step

This example is illustrative and shows how the pillars connect. The numbers below were worked by hand for teaching purposes; no code output is claimed.

Suppose three data points: x = [1, 2, 3] and y = [2, 4, 6]. The model is ŷ = w·x with a single weight, starting at w = 0, and the loss is mean squared error.

1. Linear algebra: treat x and y as vectors. Predictions are w times x, so at w = 0 the prediction vector is [0, 0, 0].
2. Calculus: the gradient of the loss with respect to w is (2/n) · Σ (w·xᵢ − yᵢ)·xᵢ.
3. Arithmetic: at w = 0 each term is −yᵢ·xᵢ, giving −2, −8 and −18, which sum to −28. The gradient is (2/3)(−28) ≈ −18.67.
4. Update: with a learning rate of 0.1, w becomes 0 − 0.1 × (−18.67) ≈ 1.87.
5. Interpretation: the true weight is 2, so one step moved w most of the way. A learning rate that is too large would overshoot and diverge, which is why a learning rate is tuned rather than guessed.

The same logic in Python is a few lines:

import numpy as np

x = np.array([1, 2, 3])
y = np.array([2, 4, 6])
w, lr = 0.0, 0.1

grad = (2 / len(x)) * np.sum((w * x - y) * x)
w = w - lr * grad
print(w)

If you can explain each line of this example in plain words, you have the core of what most beginner AI tasks require.

A six-week preparation plan

Studying one hour on weekdays and two on weekends fits alongside a normal semester timetable. Adjust the pace to your internal-assessment calendar.

Mistakes that cost beginners time

• Watching lectures without writing code. Type every example yourself.
• Jumping to neural networks before a plain baseline model has been trained.
• Reporting accuracy alone on imbalanced data.
• Fitting a scaler on the whole dataset before splitting.
• Leaving a notebook that only runs in the order you happened to click the cells.

PSG Tech and the AI prerequisites: matching this roadmap to your 2023-scheme syllabus

For students at PSG College of Technology, the most useful starting point is the official curriculum paperwork. A document from the 67th Academic Council meeting says the council considered regulations, courses of study, scheme of assessment and syllabi for BE/BTech regular programmes under the 2023 regulations (Choice Based Credit System). That document is an excerpt, so check the current version for your own programme. If your batch follows these regulations, the syllabus for your department is where the linear algebra, probability, statistics and Python rows of the scorecard above should be matched to named courses.

The same excerpt mentions industrial training, mini project/internship, inplant training and project work as components of the 2023 scheme. It also says courses such as laboratory work and industrial training are assessed for 100 marks, split into continuous assessment 40 and final evaluation 60. The excerpt is fragmentary. Credits, semester and duration are not confirmed by it, and it does not say how an online or external AI internship would be treated. Do not assume that a two-week self-directed internship fits any of these components.

In practice, this suggests a few habits. Since continuous assessment is part of how training-type courses are evaluated, keep dated evidence of your preparation: notebooks, a short GitHub repository, and weekly notes from the six-week plan. Whether any of it can be shown to a department is for the college to decide, so ask before relying on it. Treat the mathematics and programming courses in your own scheme as the base for Pillars 1 to 3, and note which machine learning topics your syllabus leaves for later semesters so that Pillar 4 practice can fill the gap.

Checks to run before you plan around PSG Tech rules

• Find the regulation year printed on your own curriculum and syllabus document. Do not assume the 2023 scheme applies to your batch.
• Ask your department office which course codes cover linear algebra, probability and statistics, Python or programming, and any machine learning subject, and in which semester each appears.
• Ask the faculty member handling internships or inplant training whether an external online internship can be recorded, and what proof of completion is accepted. Ask for the reply in writing.
• Ask what duration, timing and reporting format apply to industrial training or mini project/internship under your regulation.
• Ask whether the placement or career office needs to know about outside internships, and whether any form is needed.

Showing your preparation

When you finish the plan, you will have three things worth showing: a GitHub repository, a notebook that reproduces from top to bottom, and a short written summary of one model and its limits. That set is more convincing in an application or interview than a list of courses watched. When describing your work, state what dataset you used, what baseline you compared against and what you would do next with more time.

Frequently Asked Questions

Do I need to finish the full linear algebra and probability courses before applying?

No. The scorecard above is the practical bar: dot products, matrix shapes, conditional probability, Bayes’ theorem, and mean and variance. Deeper topics such as eigen-decomposition can follow later.

Is Python enough, or do I also need C++ or Java?

For beginner AI tasks, Python with NumPy, pandas and scikit-learn is the working standard. Other languages matter for specific roles but are not needed to complete a short guided AI project.

Can I prepare if I have not yet studied machine learning in my college courses?

Yes. The ML basics section assumes no prior course. Start with a baseline model on a small dataset and focus on the workflow and metrics rather than on algorithms.

How much calculus do I really need?

Enough to understand a derivative as a slope, compute simple partial derivatives, and follow why gradient descent moves against the gradient. Libraries calculate gradients for you, but you should be able to read what they are doing.

Should I learn deep learning first because it is popular?

Not first. A baseline built with logistic regression or a decision tree teaches data handling and evaluation, which deep learning also depends on. Move to neural networks once you can explain overfitting and data leakage.

Where can I get free datasets for practice?

The UCI Machine Learning Repository, Kaggle datasets and the sample datasets bundled with scikit-learn are all free to use. Check the licence of any dataset before publishing derived work.

What does the CareerSwami internship include as a PSG College of Technology student?

It is a free two-week (14-day) online internship with a free offer letter and a free completion certificate after successful completion, across 26 tracks. Check the track details on the CareerSwami site before you apply.

Will an online internship count for my college requirements?

That depends on your department and regulation. Use the verification list above to ask the internship coordinator, and keep their reply for your records.

Related Artificial Intelligence guides

• How Thiagarajar College of Engineering Students Can Showcase AI Internship Work in a Portfolio
• Preparing for AI Internship Interviews: A Guide for Kumaraguru College of Technology Students
• Choosing an AI Internship Track: NLP, Vision or GenAI for Vishwakarma Institute of Technology, Pune Students
• All Artificial Intelligence internship guides