GOOGLE AI-ML VIRTUAL INTERNSHIP

An Internship Report
submitted in partial fulfillment of the requirements for the award of the degree
of BACHELOR OF TECHNOLOGY
IN
COMPUTER SCIENCE & ENGINEERING
Submitted by

KOKKILIGADDA PAVAN KUMAR

21761A05G2

DEPARTMENT OF COMPUTER SCIENCE AND

ENGINEERING

LAKIREDDY BALIREDDY COLLEGE OF

ENGINEERING
(AUTONOMOUS)
Accredited by NAAC with ‘A’ Grade & NBA (Under Tier - I)
An ISO 21001:2018, 14001:2015, 50001:2018 Certified Institution
Approved by AICTE, New Delhi and Affiliated to JNTUK, Kakinada
L.B. REDDY NAGAR, MYLAVARAM, NTR Dist., ANDHRA PRADESH – 521230
2024-2025
LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING
(An Autonomous Institution since 2010) Accredited by NAAC with ‘A’ Grade & NBA (Under Tier - I),
An ISO 21001:2018, 14001:2015, 50001:2018 Certified Institution Approved by AICTE, New Delhi and Affiliated
to JNTUK, Kakinada
L.B. REDDY NAGAR, MYLAVARAM, NTR DIST., A.P.-521 230.
Phone: 08659-222 933, Fax: 08659-222931
DEPARTMENT OF COMPUTER SCIENCE & ENGINEERING

CERTIFICATE

This is to certify that internship work entitled “Google AI-ML Virtual Internship” the bonafide

work done by KOKKILIGADDA PAVAN KUMAR(21761A05G2) in partial fulfillment of the

requirements for the award of the degree in BACHELOR OF TECHNOLOGY in Computer

Science and Engineering from LAKIREDDY BALI REDDY COLLEGE OFENGINEERING,

MYLAVARAM during the year 2024-2025.

D. Anil kumar Dr. D. Veeraiah

(Internship Coordinators) (Head of the Department)

EXTERNAL EXAMINER
CERTIFICATE OF COMPLETION
 ACKNOWLEDGEMENT

I express my thanks to the support given by management in completing my work. I also

express my sincere gratitude & deep sense of respect to the Principal Dr. K. APPA RAO for
making us available all the required assistance & for his support & inspiration to carry out this
work in the Institute.

I also take the privilege to record my thanks to Dr. D. VEERAIAH Head of the
Department of CSE whose encouragement, cooperation and valuable support crown our success.

I would like to thank Mr. P. VEERA SWAMY, Sr. Assistant Professor who has been an
inspiring guide and committed faculty and gave relief moral support in every situation of
engineering career. The encouragement and support by him, especially in carrying out this project
motivated us to complete this project.
I owe my acknowledgement to an equally long list of each people who helped me in this
work. My sincere thanks to the Librarian of LBRCE who help me in getting many valuable books
of actual editors.

I am thankful to the teaching and non-teaching staff of CSE department for their direct as
well as indirect help in my work. I am elated to avail my selves to this opportunity to express my
deep sense of gratitude to my parents.

K.Pavan Kumar
21761A05G2
 DECLARATION

We hereby declare that the project report entitled “Google AI-ML Virtual Internship” is
submitted to JNTUK is partial fulfillment of the requirement forthe award of the degree of bachelor
of technology (B.Tech) is an original work carried out by us. The matter embodied in this
Internship report is a genuine work by the students and has not been submitted earlier to this
university or any other university for award of any degree or diploma or prize.

SUBMITTED BY
K.Pavan Kumar
 Executive Summary

This report summarizes my two-month AI/ML virtual internship with Google, where I gained
expertise in foundational and advanced artificial intelligence and machine learning techniques,
particularly using TensorFlow. Through practical assignments, I deepened my understanding of
AI/ML concepts and applications. A major project involved creating a computer vision model to
classify clothing types, which required developing and training convolutional neural networks
(CNNs) on diverse image datasets. This work introduced me to convolutional operations, model
optimization, and the complexities of processing visual data, alongside learning to scale models for
large datasets—an essential skill for real-world applications. Collaborating with mentors, I received
valuable feedback that refined my approach and technical skills. By the end of the internship, I had
developed strong proficiency in TensorFlow, image processing, and neural network implementation,
equipping me with skills and insights for future work in AI/ML and laying a solid foundation for
further exploration in the field.
 Table of Contents

Name of the Content Pg.No

1. Introduction 2

2. Internship Activities 3

2.1 Role and Responsibilities

2.2 Projects Undertaken

3. Skills and Knowledge Gained 11

4. Challenges and How I Overcame Them 13

5. Conclusion 15

6. Recommendations 16

7. Learning Outcomes 16

8. Appendices 18

9. References 20
1. Introduction:

Objective of the Internship:

The primary objective of this AI/ML virtual internship with Google, under the support of AICTE
Eduskills and India Edu, was to build practical skills in artificial intelligence and machine learning
with a focus on computer vision. Through this internship, I aimed to gain hands-on experience
with foundational and advanced ML techniques, particularly using TensorFlow, to understand the
complete machine learning pipeline—from building basic models to implementing complex neural
networks for real-world applications.

Company Overview:

Google, a leader in technology and AI innovation, provided the framework for this internship.
Known for pioneering advancements in artificial intelligence, machine learning, and data
processing, Google has developed a robust platform for ML practitioners, including TensorFlow,
which is one of the most widely used libraries for machine learning. Google’s commitment to
supporting emerging talent in AI/ML through programs like this internship aligns with their goal
to democratize AI technology and build a community of skilled practitioners.

Relevance to my field:

As AI and machine learning continue to shape the future of technology, this internship offered a
critical opportunity to build expertise directly applicable to my academic and professional goals.
By focusing on TensorFlow, I engaged with one of the most powerful tools for AI development,
acquiring skills in neural networks, computer vision, and data processing—all highly relevant to
my career aspirations in data science and artificial intelligence. The skills gained from this
experience not only strengthened my technical foundation but also prepared me for tackling
complex machine learning challenges in both academic research and industry settings.

1
2. Internship Activities

2.1 Roles and Responsibilities:

1) Building and Training Machine Learning Models: Developed and trained foundational
ML models using TensorFlow, progressing from simple models to complex architectures for
real-world applications, particularly in computer vision.

2) Implementing Neural Networks for Computer Vision: Worked on projects that involved
building convolutional neural networks (CNNs) to classify images, including identifying
various types of clothing and distinguishing between images of humans and animals.

3) Using Convolutions and Pooling Layers: Applied convolution and pooling techniques to
improve model performance, focusing on feature extraction and enhancing model accuracy
when handling complex visual data.

4) Working with Large Datasets: Managed and processed large image datasets to train
models effectively, learning to address issues like overfitting and optimizing models for higher

5) Using Cloud-Based Tools (Google Colab): Leveraged Google Colab for model
development and training, allowing access to GPU resources, necessary libraries, and
collaborative tools for enhanced productivity and code sharing.

6) Exploring Data Visualization Techniques: Used TensorBoard to visualize model metrics

and performance, helping to analyze training progress, loss functions, and accuracy over time,
which provided insights for model tuning.

7) Implementing Model Optimization Techniques: Focused on enhancing model accuracy

by experimenting with different network layers and regularization techniques to prevent
overfitting, making models more generalizable.

2
2.2 Projects Undertaken
Project Title 1: Clothing Recognition Using Convolutional Neural

Networks Introduction:

With the growing demand for intelligent visual recognition systems, computer vision applications
have become vital across various industries. This project focused on developing a machine
learning model capable of identifying different articles of clothing through image recognition.
Using TensorFlow and convolutional neural networks (CNNs), the model was trained on a dataset
of clothing images, allowing it to learn and distinguish patterns, shapes, and textures associated
with different clothing types. Through this experience, I gained practical skills in model
development, image preprocessing, and neural network training, equipping me to contribute to
further advancements in AI-driven computer vision applications.

Literature Survey
1) Clothing Recognition Using Machine Learning:

Using machine learning models like Convolutional Neural Networks (CNNs), clothing
recognition systems can identify different types of clothing in images. CNNs are widely used for
image tasks because they can detect unique patterns, textures, and shapes, making them effective
for recognizing various clothing items in diverse settings.
2) Image data processing and avoid overfitting:
Preprocessing techniques like resizing, normalization, and data augmentation (e.g.,
rotating or flipping images) help prepare large image datasets for better model training. These
steps improve the model's accuracy and make it more adaptable to new, unseen data. Techniques
like dropout and data augmentation, along with using diverse datasets, help prevent overfitting and
ensure the model generalizes well.
3) Cloud Tools for Model Training:
Platforms like Google Colab provide the computing power needed for training large
models. Visualization tools like TensorBoard help track training progress, making it easier to
improve model performance.

3
Methodology
1. Data Collection and Preprocessing
 Dataset Selection: A clothing image dataset was chosen, containing various categories of
clothing items.
 Data Preprocessing: Each image was resized, normalized, and augmented (using techniques
like rotation and flipping) to increase diversity and improve model robustness.
 Data Splitting: The dataset was split into training, validation, and test sets to ensure reliable
evaluation of the model’s performance.

Model Architecture and Training

 Building the CNN: A convolutional neural network (CNN) was designed with layers
for feature extraction (convolution and pooling) and classification (fully connected
layers).
 Training: The CNN model was trained on the preprocessed images using
TensorFlow, optimizing weights to identify patterns unique to each clothing type.
 Hyperparameter Tuning: Learning rate, batch size, and number of epochs were adjusted to
improve accuracy and training efficiency.
4
Model Optimization and Evaluation

 Avoiding Overfitting: Techniques like dropout and regularization were applied to

reduce overfitting and improve generalization.

 Performance Metrics: Metrics such as accuracy, precision, and recall were used to
evaluate the model’s performance on the test dataset.

 Visualization: Training and validation metrics were visualized using TensorBoard to monitor
the model's learning process and make adjustments as needed.

Deployment and Testing

 Cloud Deployment: The final model was deployed on Google Colab, making use of
its computational resources.
 Real-World Testing: The model was tested on additional images to assess its practical
usability and accuracy in identifying clothing items.

5
Project Title 2: Use convolutional neural networks (CNNs)
with complex images
Introduction
This project focuses on building a convolutional neural network (CNN) model to classify
images of horses and humans, leveraging the power of CNNs to distinguish between these
visually distinct classes. Using TensorFlow, a widely adopted framework for machine learning,
this project aims to train and deploy a model capable of accurately identifying whether an image
contains a horse or a human. By developing a robust image classification model, this project
contributes to the growing need for automated recognition systems that can assist in diverse
scenarios, from identifying animals in wildlife monitoring to supporting intelligent video analysis
in other applications.

6
Literature Survey
1. Image classifications with CNNs:

CNNs have become a standard in image classification due to their ability to extract and learn
spatial hierarchies of features from images.Studies have shown that CNNs outperform traditional
machine learning models in accuracy and efficiency for image classification tasks.
2. Feature Extraction and Pooling
Convolutional and pooling layers form the backbone of CNNs, allowing models to detect
patterns in images. Convolutional layers capture essential features like edges and textures, while
pooling layers reduce the spatial dimensions, making the model more computationally efficient.
3. Data Augmentation for Improved Generalization
Data augmentation techniques, such as rotating, flipping, and scaling, have proven
effective in expanding limited datasets and reducing overfitting. For small datasets, augmentation
has been shown to improve model generalization by artificially increasing the data variability.
4. Overfitting and Regularization Techniques
Overfitting is a common challenge in training deep learning models on limited data.
Techniques like dropout and batch normalization are frequently employed to mitigate overfitting
by adding noise to the training process and ensuring stable learning.

Methodology
1) Data Collection and Preprocessing

o Dataset Selection: A dataset of labeled images of horses and humans was chosen to train
the classification model.
o Data Preprocessing: Images were resized and normalized to ensure consistency. Data
augmentation techniques, such as flipping and rotating, were applied to increase data
variety and enhance the model’s ability to generalize to new images.
o Data Splitting: The dataset was divided into training, validation, and test sets to enable
accurate model evaluation.

7
2) Model Architecture and Training

o Building the CNN: A convolutional neural network (CNN) was built with convolutional
and pooling layers to extract image features and fully connected layers for classification.
o Training: The model was trained using TensorFlow, optimizing it to recognize and
distinguish features specific to horses and humans.
o Hyperparameter Tuning: Parameters like batch size, learning rate, and the number of
epochs were adjusted to improve accuracy and performance.

3) Model Optimization and Evaluation

o Avoiding Overfitting: Regularization techniques like dropout were applied to reduce

overfitting and improve the model’s generalization ability.
o Performance Metrics: Metrics such as accuracy, precision, and recall were used to
evaluate the model’s performance on the test set.
o Visualization: Training and validation metrics were monitored using TensorBoard to
observe learning progress and fine-tune the model as necessary.
4) Deployment and Testing

o Cloud Deployment: The model was deployed and tested in Google Colab, leveraging its
computational resources.
o Real-World Testing: Additional images were used to test the model’s practical ability to
classify images of horses and humans, evaluating its usability and accuracy.
8
 Classified as Human:

Classified as Horse:

9
3. Skills and Knowledge Gained
During my virtual internship, I developed a range of technical and soft skills essential to the field
of AI and machine learning:
Technical Skills:
• Machine Learning Fundamentals: Developed a solid foundation in core machine
learning principles, covering various algorithms and approaches to solve classification
and prediction tasks. I became proficient in understanding and applying different models
based on specific project requirements.
• Convolutional Neural Networks (CNNs): Built and optimized CNNs to classify images,
gaining in-depth knowledge of convolutional operations, pooling layers, and feature
extraction techniques. Through hands-on practice, I learned how CNNs are structured to
recognize patterns in images and how to fine-tune them for improved accuracy.
• Computer Vision: Specialized in applying CNNs for computer vision tasks. I worked on
projects that involved image classification and object recognition, learning techniques to
enhance model performance on complex image data and recognizing features even in
diverse and cluttered scenes.
• Deep Learning with TensorFlow: Acquired practical experience with TensorFlow and
Google Colab as my primary tools for model building and experimentation. I became
adept at using Colab's GPU acceleration and cloud-based resources, which enabled
efficient training and testing of machine learning models, particularly when working with
larger datasets.
• Model Optimization Learned advanced techniques to improve model performance and
avoid overfitting, such as data augmentation, dropout, and regularization. This helped me
design models that generalize better on unseen data, essential for achieving real-world
application readiness.
• Data Handling and Large Dataset Management: Gained insights into managing large
datasets efficiently, including techniques for loading, preprocessing, and batching data to
optimize training time and memory usage. This experience is crucial for working on
scalable machine learning solutions.
• Cloud-Based Deployment: Used Google Colab to deploy and test models, leveraging
cloud resources for scalable and efficient training processes.
10
 • Hyperparameter Tuning: Developed skills in hyperparameter tuning to enhance model
accuracy. I practiced adjusting parameters like learning rate, batch size, and layer
configurations to optimize model performance while maintaining computational
efficiency.
• Data Augmentation and Preprocessing: Acquired skills in augmenting data to improve
model generalizability. Techniques like flipping, rotating, and resizing images were
applied to increase dataset variability, which in turn enhanced the robustness of the
trained models.
• Programming: Enhanced skills in Python, focusing on TensorFlow and other essential
libraries, such as NumPy, for machine learning workflows.
Soft Skills:
• Communication: Strengthened communication skills through interactions with mentors,
actively receiving and applying constructive feedback on projects.
• Time Management: Managed multiple codelabs and assignments within deadlines, honing
my ability to prioritize tasks in a structured learning environment.
• Problem-Solving: Developed a problem-solving mindset, identifying and addressing
challenges in model development, from debugging code to fine-tuning model parameters.
• Adaptability: Improved adaptability by quickly adjusting to new tools, methodologies, and
concepts in machine learning and AI as the internship progressed.
• Self-Motivation and Initiative: Cultivated a proactive approach to learning by
independently exploring advanced concepts and seeking additional resources to reinforce
my understanding.

11
4. Challenges and How I Overcame Them
During my virtual AI/ML internship, I faced several challenges that helped me grow both
technically and professionally. Below are the key challenges and the steps I took to overcome
them:
1. Challenge: Understanding Complex Machine Learning Concepts
• Description: Initially, I found some advanced machine learning concepts, especially
convolutional neural networks (CNNs) and their applications in computer vision, to be
challenging. The theoretical aspects of CNNs and understanding how various layers (e.g.,
convolution, pooling, fully connected layers) contribute to feature extraction and
classification were difficult to grasp at first.
• How I Overcame It: To overcome this challenge, I focused on breaking down the
concepts into smaller, more digestible parts. I used various online resources, such as
tutorials, blog posts, and videos, to clarify my understanding. Additionally, I experimented
hands-on with code examples and gradually built more complex models. This hands-on
approach helped me internalize the concepts and improve my comprehension.
2. Challenge: Model Optimization and Overfitting
• Description: A significant challenge during my internship was optimizing machine
learning models, particularly dealing with overfitting. My early models performed well on
training data but didn’t generalize effectively to unseen data, which is crucial for real-
world applications.
• How I Overcame It: I addressed this issue by experimenting with various regularization
techniques, such as dropout, L2 regularization, and data augmentation. I also learned how
to monitor the validation performance during training to ensure that the model wasn’t
overfitting. By tweaking hyperparameters and implementing these techniques, I was able to
improve model generalization and performance.
3. Challenge: Handling Large Datasets
• Description: Training models on large image datasets presented a challenge in terms of
computational resources and processing time. The models were slow to train, and I
frequently encountered memory issues, particularly when running experiments on local
machines with limited GPU resources.

12
 • How I Overcame It: I leveraged cloud computing resources provided by Google Colab,
which offered free access to GPUs and TPUs. This allowed me to run experiments more
efficiently and train models faster. Additionally, I optimized the dataset by using image
resizing, batching, and caching techniques to speed up data loading and reduce memory
usage.
4. Challenge: Experimenting with Different Architectures
• Description: Trying to build the most accurate model meant experimenting with different
architectures and layers. The process of selecting the right combination of layers, neurons,
and activation functions for a CNN model was time-consuming and required a lot of trial
and error.
• How I Overcame It: I adopted an iterative approach to experimentation, where I
systematically adjusted the architecture based on the results from the previous tests. I used
resources like research papers and previous models to understand which architectures
typically work best for specific types of tasks. This allowed me to fine-tune my models and
achieve better performance.
5. Challenge: Debugging and Model Validation
• Description: Debugging machine learning models to identify the cause of poor
performance or errors was another challenge. At times, the models wouldn’t behave as
expected, and figuring out where things went wrong in the data pipeline or model layers
was not straightforward.
• How I Overcame It: To tackle this, I used systematic debugging techniques like checking
intermediate outputs at various stages of the pipeline. I also compared my model's
predictions against the ground truth to identify any patterns in the errors. I frequently
sought feedback from mentors to gain new perspectives on the issues. This helped me
identify and resolve the root causes of the problems.

13
5. Conclusion:
My virtual internship focused on AI/ML and image classification was an invaluable
experience that allowed me to deepen my understanding of cutting-edge technologies like machine
learning, computer vision, and deep learning. It provided me with a platform to work on hands-on
projects, such as building and training convolutional neural networks (CNNs) to classify images of
horses and humans. This allowed me to apply theoretical concepts from my academic background
to real-world applications, enhancing both my technical and problem-solving abilities.
Throughout the internship, I gained technical expertise in areas such as:
• Machine Learning Model Development: I learned how to develop, train, and optimize
models, particularly CNNs, to classify images and improve accuracy by implementing
techniques like data augmentation and regularization.
• Model Evaluation and Debugging: I enhanced my skills in model validation, debugging,
and fine-tuning to improve the accuracy of predictions.
• Practical Machine Learning Implementation: I worked on data preprocessing, model
training, and evaluation, which solidified my understanding of real-world machine learning
workflows.
In addition to the technical knowledge, this internship strengthened my soft skills, such as
communication, time management, and collaboration.Overall, this internship has not only
improved my technical skills but has also given me a deeper insight into the real-world
applications of AI/ML in image classification I feel confident that the knowledge gained during
this internship will be helpful for my future career in machine learning, AI, and software
development.

14
 6. Recommendations:
For future interns pursuing an AI/ML-focused internship or working on image classification
projects, I recommend deepening your understanding of machine learning fundamentals,
particularly around neural networks and convolutional neural networks (CNNs). It’s essential to
grasp concepts like overfitting, underfitting, and model evaluation metrics, as these will
significantly enhance your learning experience. Familiarizing yourself with TensorFlow and tools
like Google Colab will also prove beneficial, as these are commonly used for deep learning tasks
and cloud-based computing. Additionally, engaging in active collaboration with mentors and peers
is crucial. Regularly seeking feedback from others will help refine your approach to problem-
solving, improve model performance, and introduce new techniques. It’s also important to focus
on data preprocessing—understanding methods such as data augmentation, normalization, and
how to handle unbalanced data will improve model accuracy. Staying proactive in learning new
tools and techniques and asking questions when encountering challenges will contribute to your
growth. Applying the concepts learned to real-world problems will help reinforce theoretical
knowledge, making the experience more impactful. By following these recommendations, future
interns will be able to gain a comprehensive understanding of machine learning and AI, build
practical skills, and maximize their internship experience.

7. Learning Outcomes

1. Enhanced Technical Skills:

o Gained hands-on experience in developing machine learning models using TensorFlow
for computer vision tasks, including working with convolutional neural networks
(CNNs).
o Improved understanding of model training, optimization, and deployment for image
classification tasks.
o Developed expertise in using cloud-based platforms like Colab for model development
and testing, facilitating real-time collaboration and efficient coding.

15
2. Industry-Specific Knowledge:
o Gained exposure to modern AI/ML techniques in image recognition, including
convolutional operations and feature map creation.
o Learned to apply machine learning principles to real-world image data, including the
classification of clothing items and distinguishing between horses and humans in images.
o Familiarized with the practical use of machine learning models for real-world
applications in AI/ML environments, emphasizing scalability and performance.
3. Problem-Solving and Analytical Skills:
o Addressed challenges such as improving model accuracy, managing dataset complexities,
and optimizing performance in the context of computer vision models.
o Collaborated with mentors and peers to troubleshoot issues and refine the models,
enhancing problem-solving and analytical skills.
o Applied critical thinking to enhance model generalization and overcome overfitting
through the use of large datasets.
4. Time Management and Prioritization:
o Managed multiple tasks including model training, experimentation, and optimization in a
timely manner.
o Developed strong prioritization skills to balance project deadlines and experimentation
within the internship timeframe.
o Worked efficiently on different model layers and image classification tasks while meeting
deadlines.
5. Improved Communication and Teamwork:
o Participated in regular team meetings and code reviews, significantly improving
communication skills and collaborative problem-solving.
o Enhanced my ability to communicate technical details effectively through presentations
and discussions with peers and mentors.
o Fostered teamwork by working on shared goals and incorporating feedback to improve
model development.

16
6. Adaptability to Professional Work Environment:
o Adapted to a virtual work environment by learning to use collaborative tools like Google
Colab, version control systems (Git), and online documentation.
o Quickly became comfortable with using cloud platforms and tools for large-scale data
processing and model deployment.
o Adjusted to dynamic work tasks, including changes in project focus and technical
challenges that arose during the internship.
7. Career Insights:
o Gained practical experience with TensorFlow and AI/ML techniques, strengthening my
interest in pursuing a career in machine learning and artificial intelligence.
o Gained a deeper understanding of how AI/ML models are applied in real-world scenarios
like image classification and computer vision.
o Interacted with professionals in the AI/ML field, which reinforced my career aspirations
and provided guidance on the skills necessary for success in AI/ML roles.

8. Appendices:
Project 1:
Preparing the model:

17
Project 2:
Preparing the model:

18
19
9.References:
1. https://developers.google.com/learn/pathways/tensorflow#codelabhttps://developers.google.co

m/codelabs/tensorflow-5-compleximages
2. https://machinelearningmastery.com/how-to-develop-a-cnn-from-scratch-for-fashion-mnist-
clothing-classification/
3. https://developers.google.com/codelabs/tensorflow-5- compleximages?continue=https%3A
%2F%2Fdevelopers.google.com%2Flearn%2Fpathways%2 Ftensorflow%23codelab-https%3A
%2F%2Fdevelopers.google.com%2Fcodelabs%2Ftensorflow- 5-compleximages#0
4. https://www.kaggle.com/code/dskagglemt/horse-or-human-using-cnn
5. https://github.com/Billl-11/CNN-Horse-vs-Human-Classifier
6. https://openai.com/index/chatgpt/

20