Python Developer (AI/ML) fresher role

1. Python Programming (Core)

1. Q: What are Python’s key features?

A: Python is interpreted, dynamically typed, supports object-oriented programming, has extensive standard
libraries, and emphasizes readability and simplicity.

2. Q: Difference between lists and tuples?

A: Lists are mutable, tuples are immutable; lists use [], tuples use (). Tuples are faster for read-only operations.

3. Q: Explain Python decorators.

A: Decorators are functions that modify other functions without changing their code. Example: @staticmethod or
custom decorators using def wrapper(func): ....

4. Q: Difference between deepcopy and shallow copy?

A: Shallow copy copies the object but references nested objects; deepcopy copies the object and all nested objects
recursively.

5. Q: How does Python handle memory management?

A: Python uses reference counting and a garbage collector for cyclic references to manage memory automatically.

6. Q: Explain Python’s GIL.

A: Global Interpreter Lock (GIL) allows only one thread to execute Python bytecode at a time, limiting true
parallelism in CPU-bound threads.

7. Q: How do you handle exceptions in Python?

A: Using try-except blocks. Optionally, finally for cleanup and else for code when no exception occurs.

8. Q: Difference between Python is and ==?

A: is checks object identity (memory location), == checks value equality.

9. Q: What are Python generators?

A: Generators yield items one at a time using yield, saving memory for large sequences.

10. Q: Explain list comprehensions.

A: Compact syntax to create lists: [x**2 for x in range(10) if x%2==0].

2. AI/ML Fundamentals

1. Q: Difference between supervised and unsupervised learning?

A: Supervised uses labeled data; unsupervised finds patterns in unlabeled data.

2. Q: What is overfitting and how do you prevent it?

A: Overfitting occurs when a model fits training data too closely. Prevent using regularization, dropout, or more
data.

3. Q: Explain Isolation Forest.

A: An unsupervised algorithm for anomaly detection; it isolates anomalies via random partitioning of data.

4. Q: How does a CNN work?

A: Convolutional layers extract spatial features, pooling layers reduce dimensionality, and fully connected layers
classify features.

5. Q: Difference between RNN and LSTM?

A: RNNs suffer from vanishing gradients; LSTM adds gates to maintain long-term dependencies.
 6. Q: What is Autoencoder?
A: Neural network used for unsupervised feature learning and anomaly detection; compresses input into latent
space and reconstructs output.

7. Q: How do you evaluate a classification model?

A: Using metrics like accuracy, precision, recall, F1-score, and confusion matrix.

8. Q: Explain the train-test split.

A: Splitting data into training (to train model) and testing (to evaluate model performance) sets.

9. Q: Difference between regression and classification?

A: Regression predicts continuous values, classification predicts discrete labels.

10. Q: How do you detect anomalies in time-series data?

A: Using statistical methods, isolation forests, autoencoders, or LSTM-based prediction errors.

3. Data Analysis / Preprocessing

1. Q: How do you handle missing values?

A: Drop rows/columns, impute with mean/median/mode, or use predictive models.

2. Q: Explain one-hot encoding.

A: Converts categorical values into binary columns for ML models.

3. Q: How do you normalize or standardize data?

A: Normalization scales data between 0–1; standardization converts to zero mean and unit variance.

4. Q: What is feature engineering?

A: Creating new features or transforming existing ones to improve model performance.

5. Q: How do you detect outliers?

A: Using IQR, Z-score, boxplots, or ML algorithms like Isolation Forest.

6. Q: Difference between NumPy array and Python list?

A: NumPy arrays are homogeneous, support vectorized operations, and are memory efficient.

7. Q: How do you merge/join datasets in Pandas?

A: Using merge(), concat(), join() based on keys or indexes.

8. Q: How to handle categorical data in ML?

A: Encoding: one-hot, label encoding, target encoding.

9. Q: Explain rolling/window functions in Pandas.

A: Used for moving averages, sums, or custom operations over time-series windows.

10. Q: How do you visualize data distributions?

A: Using histograms, boxplots, KDE plots with Matplotlib/Seaborn/Plotly.

4. Projects / Practical Knowledge

1. Q: Explain your CCTV optimization project.

A: I preprocessed CCTV footage, applied YOLOv4 for object detection, used Deep SORT for tracking, and applied
CNN anomaly detection to select key frames, reducing storage.

2. Q: How did you integrate GenAI in the CCTV project?

A: Used LangChain and GPT/Claude LLMs to query surveillance events via natural language interface.

3. Q: Challenges faced in lane detection project?

A: Handling shadows, varying lighting, occlusions; solved using Hough transform, edge detection, and robust
preprocessing.
 4. Q: Which metrics did you use to evaluate anomaly detection?
A: Precision, recall, F1-score, and detection accuracy on known anomalies.

5. Q: How did you optimize data pipelines?

A: Vectorized operations with NumPy/Pandas, avoided loops, used batch processing for large video frames.

6. Q: How did you handle imbalanced data?

A: Oversampling, undersampling, or using anomaly detection algorithms suited for rare events.

7. Q: Which deep learning frameworks did you use?

A: TensorFlow, Keras, PyTorch for Autoencoders, CNNs, and anomaly detection.

8. Q: How do you deploy models in production?

A: Using Flask APIs, saving model artifacts, handling inference requests.

9. Q: How did you evaluate YOLOv4 performance?

A: Using mAP (mean Average Precision) and tracking consistency across frames.

10. Q: How did you handle real-time video data?

A: Batch processing, streaming frames with OpenCV, reducing computational load with frame skipping and anomaly
selection.

5. Cloud / Deployment

1. Q: How do you deploy a Python ML model?

A: Save model using Pickle or joblib, create a REST API using Flask, handle requests and return predictions.

2. Q: How do you scale ML apps in the cloud?

A: Use AWS services (EC2, Lambda), containerization (Docker), and auto-scaling for load management.

3. Q: What is AWS S3 used for?

A: Storing large datasets, model artifacts, and outputs securely with high availability.

4. Q: Difference between AWS EC2 and Lambda?

A: EC2 provides virtual machines; Lambda is serverless, runs code on demand.

5. Q: How do you secure ML models on cloud?

A: IAM policies, encryption at rest and transit, access control, and API authentication.

6. Q: Explain how to integrate a model with a Flask API.

A: Load model on server start, define routes for prediction, process input, return JSON responses.

7. Q: What’s the benefit of containerization?

A: Ensures consistency across environments, easier deployment, and scalability.

8. Q: How do you handle large datasets in AWS?

A: Use S3 for storage, EC2 or SageMaker for processing, batch or stream processing depending on need.

9. Q: What is inference latency and why does it matter?

A: Time taken for model to return predictions; affects real-time systems like CCTV monitoring.

10. Q: How do you monitor deployed ML models?

A: Logging predictions, tracking performance metrics, retraining if model drifts.

6. AI/ML Tools / Libraries

1. Q: Why use OpenCV?

A: For image/video processing, object detection, feature extraction, and computer vision tasks.
 2. Q: Difference between TensorFlow and PyTorch?
A: Tensor

Flow uses static graphs (eager execution optional), PyTorch uses dynamic graphs; PyTorch is more Pythonic and flexible for
research.

3. Q: What is scikit-learn used for?

A: Traditional ML algorithms, preprocessing, model evaluation, and pipeline creation.

4. Q: What is LangChain?
A: Python framework to build LLM-powered applications with prompts, chains, and agents.

5. Q: How do you perform data augmentation?

A: Rotate, flip, crop, add noise, scale images to increase dataset diversity.

6. Q: Why use YOLOv4?

A: Real-time object detection with high accuracy and speed.

7. Q: What is Deep SORT?

A: Tracking algorithm combining Kalman filters and appearance descriptors for multi-object tracking.

8. Q: Difference between TensorFlow and Keras?

A: Keras is high-level API for building neural networks; TensorFlow is the backend engine.

9. Q: How do you use Pandas for large datasets?

A: Chunking, vectorized operations, memory optimization, and efficient file formats (Parquet/Feather).

10. Q: How do you visualize anomalies in Python?

A: Use Matplotlib/Seaborn to plot trends, scatter plots, heatmaps, or dashboard visualization with Plotly/Power BI.

ALL possible coding questions with answers:

1. Python Basics (10 Q&A)

Q1: Reverse a string in Python.

A1:

s = "V2Soft"

reversed_s = s[::-1]

print(reversed_s) # tofs2V

Q2: Check if a number is prime.

A2:

def is_prime(n):

if n <= 1: return False

for i in range(2, int(n**0.5)+1):

if n % i == 0:

return False

return True

print(is_prime(17)) # True
Q3: Count frequency of elements in a list.
A3:

from collections import Counter

lst = [1,2,2,3,3,3]

print(Counter(lst)) # {1:1, 2:2, 3:3}

Q4: Swap two variables without a temp variable.

A4:

a, b = 5, 10

a, b = b, a

print(a, b) # 10 5

Q5: Find the largest number in a list.

A5:

lst = [3,9,1,7]

print(max(lst)) # 9

Q6: List comprehension: squares of even numbers from 1-10.

A6:

squares = [x**2 for x in range(1,11) if x%2==0]

print(squares) # [4,16,36,64,100]

Q7: Merge two dictionaries.

A7:

d1 = {'a':1}; d2={'b':2}

merged = {**d1, **d2}

print(merged) # {'a':1,'b':2}

Q8: Flatten a nested list.

A8:

lst = [[1,2],[3,4]]

flat = [item for sub in lst for item in sub]

print(flat) # [1,2,3,4]

Q9: Check if a string is a palindrome.

A9:

def is_palindrome(s):

return s == s[::-1]

print(is_palindrome("radar")) # True

Q10: Remove duplicates from a list.

A10:

lst = [1,2,2,3]

unique = list(set(lst))

print(unique) # [1,2,3]
2. Data Structures & Algorithms (10 Q&A)

Q1: Implement a stack using a list.

A1:

stack = []

[Link](1)

[Link](2)

print([Link]()) # 2

Q2: Reverse a linked list (Python example with class).

A2:

class Node:

def __init__(self, val):

[Link] = val

[Link] = None

def reverse(head):

prev, curr = None, head

while curr:

nxt = [Link]

[Link] = prev

prev = curr

curr = nxt

return prev

Q3: Find the middle element of a list.

A3:

lst = [1,2,3,4,5]

middle = lst[len(lst)//2]

print(middle) # 3

Q4: Implement queue using deque.

A4:

from collections import deque

q = deque()

[Link](1)

[Link](2)

print([Link]()) # 1

Q5: Binary search in sorted array.

A5:
def binary_search(arr, target):

l, r = 0, len(arr)-1

while l <= r:

m = (l+r)//2

if arr[m]==target: return m

elif arr[m]<target: l=m+1

else: r=m-1

return -1

print(binary_search([1,3,5,7],5)) # 2

Q6: Merge two sorted lists.

A6:

def merge(a,b):

res=[]

i=j=0

while i<len(a) and j<len(b):

if a[i]<b[j]: [Link](a[i]); i+=1

else: [Link](b[j]); j+=1

[Link](a[i:]); [Link](b[j:])

return res

print(merge([1,3],[2,4])) # [1,2,3,4]

Q7: Find first non-repeating character in string.

A7:

from collections import Counter

s = "swiss"

freq = Counter(s)

for c in s:

if freq[c]==1: print(c); break # w

Q8: Check if brackets are balanced.

A8:

def balanced(s):

stack=[]

d = {'(':')','[':']','{':'}'}

for c in s:

if c in d: [Link](c)

elif not stack or d[[Link]()]!=c: return False

return not stack

print(balanced("({[]})")) # True

Q9: Find kth largest element in list.

A9:

import heapq

lst=[3,2,1,5,6,4]

print([Link](1,lst)[0]) # 6

Q10: Implement simple hash map using dict.

A10:

hm = {}

hm['key'] = 100

print(hm['key']) # 100

3. Data Handling / ML Problems (10 Q&A)

Q1: Load CSV and compute mean of a column.

A1:

import pandas as pd

df = pd.read_csv("[Link]")

print(df['col'].mean())

Q2: Normalize a numpy array.

A2:

import numpy as np

arr = [Link]([1,2,3])

normalized = (arr - [Link]()) / ([Link]() - [Link]())

print(normalized) # [0,0.5,1]

Q3: Handle missing values in pandas.

A3:

[Link]([Link](), inplace=True)

Q4: Split dataset into train/test.

A4:

from sklearn.model_selection import train_test_split

X_train,X_test,y_train,y_test = train_test_split(X,y,test_size=0.2,random_state=42)

Q5: Fit a linear regression model.

A5:

from sklearn.linear_model import LinearRegression

model = LinearRegression()

[Link](X_train,y_train)

y_pred = [Link](X_test)
Q6: Evaluate model with accuracy.
A6:

from [Link] import accuracy_score

accuracy_score(y_test, y_pred.round())

Q7: One-hot encode categorical variable.

A7:

pd.get_dummies(df['category'])

Q8: Detect outliers using Isolation Forest.

A8:

from [Link] import IsolationForest

iso = IsolationForest(contamination=0.1)

outliers = iso.fit_predict(X)

Q9: Train a simple CNN using Keras.

A9:

from [Link] import Sequential

from [Link] import Conv2D, Flatten, Dense

model = Sequential([

Conv2D(32,(3,3),activation='relu',input_shape=(28,28,1)),

Flatten(),

Dense(10,activation='softmax')

])

[Link]('adam','sparse_categorical_crossentropy',metrics=['accuracy'])

Q10: Load pre-trained YOLOv4 model in Python (OpenCV).

A10:

import cv2

net = [Link]("[Link]","[Link]")