Cognizant: GenC (AI/ML Track) Interview

def is_prime(n: int) -> bool:
    if n < 2: return False
    if n % 2 == 0: return n == 2
    i = 3
    while i * i <= n:
        if n % i == 0: return False
        i += 2
    return True

• Checking divisors only up to √n gives O(√n). For many numbers at once, mention the Sieve of Eratosthenes, which is O(n log log n).

def merge_sort(a):
    if len(a) <= 1: return a
    m = len(a) // 2
    L, R = merge_sort(a[:m]), merge_sort(a[m:])
    out, i, j = [], 0, 0
    while i < len(L) and j < len(R):
        if L[i] <= R[j]: out.append(L[i]); i += 1
        else: out.append(R[j]); j += 1
    return out + L[i:] + R[j:]

• O(n log n) in every case, O(n) extra space, and stable (equal items keep their order). Quicksort averages O(n log n) and sorts in place, but its worst case is O(n²).

• Encapsulation: bundle data with the methods that use it, and hide the internals behind a clear interface.
• Abstraction: expose what an object does, not how (abstract classes and interfaces).
• Inheritance: a child class reuses and extends a parent class.
• Polymorphism: one interface, many implementations (method overriding, duck typing in Python).
• Wrapper class (Java): an object version of a primitive (int → Integer) so it can be used in collections; autoboxing converts between them. In Python the general idea is a class that wraps another object to add behaviour, such as a decorator or adapter.

• INNER JOIN returns matching rows only. LEFT JOIN returns all left rows plus matches (NULL where there is none). RIGHT JOIN is the mirror image. FULL OUTER JOIN returns all rows from both sides.

SELECT e.emp_id, e.name
FROM employees e
LEFT JOIN departments d ON d.dept_id = e.dept_id
WHERE d.dept_id IS NULL;

• This is the "anti-join" pattern. NOT EXISTS is an alternative that handles NULLs more safely than NOT IN.

Answer (a 3-minute structure):

1. Problem and data: "Predict loan default from 50k applications with 30 features."
2. Approach: cleaning (missing values, outliers), features, a baseline logistic regression, then XGBoost, with stratified cross-validation.
3. Result: "AUC 0.84 against a 0.76 baseline; recall of 0.71 at 0.8 precision."
4. Your part: "I built the feature pipeline and the Streamlit demo."
5. What you'd improve: "Calibration, SHAP explanations and deployment with FastAPI." - Expect follow-ups such as "why XGBoost?", "how did you handle imbalance?" and "what is AUC?". Know every line of your resume.

• Supervised: you learn from labelled examples (X → y).
• Classification: spam detection, churn.
• Regression: price prediction.
• Algorithms: linear and logistic regression, decision trees, random forests, SVM, neural networks.
• Unsupervised: there are no labels; you find structure in the data.
• Clustering: customer segments with k-means.
• Dimensionality reduction: PCA.
• Anomaly detection.
• Also mention: semi-supervised learning, self-supervised learning (how LLMs are pre-trained), and reinforcement learning (rewards).