Sahil Garg

AI-Engineer| Computer Vision | Machine Learning | Deep Learning |Data

Pipeline

Highly skilled Python Developer with expertise in Python, MySQL, Machine

Learning and Deep learning with 2 years of experience. Excellence in problem-
solving and data structures. Committed to delivering high-quality projects with a
focus on innovation and efficiency.
Technical Skills
Languages: Python, C++

Domain: Data Engineering, Machine Learning, LLMs

, Computer vision, Object Detection, Optical
Character recognition ,Image Segmentation,
natural language processing

Frameworks: Django, PyTorch ,TensorFlow, FastAPI,

Scikit-learn, OpenCV,Docker

CI/CD Tools: GitLab

Databases: MySQL, MongoDB, Postgres

Cloud: Azure,Aws
Operating Systems: Linux, Windows
Concepts & Processes: DevOps, CI/CD, Design Patterns, Problem
Solving
Data Engineering Tools: Azure Data Factory, Azure Synapse analytics
PowerBI ,Pyspark
Key Responsibilities and Achievements:

 Developed and maintained large-scale software products, ensuring

high performance and scalability.
 Worked extensively with big data technologies, including the
development and optimization of data ingestion pipelines using
PySpark.
 Leveraged cloud platforms such as Azure, AWS, and GCP to
deploy and manage applications, ensuring seamless integration and
continuous deployment.
 Wrote and maintained unit and integration tests to ensure the
robustness of ingestion pipelines and data processing systems.
 Implemented static analysis and code quality tools to maintain high
standards of code health and maintainability.
 Utilized DevOps principles and CI/CD pipelines to streamline
deployment processes, improving efficiency and reducing
downtime.
 Demonstrated excellent communication and leadership skills,
effectively collaborating with cross-functional teams in a fast-
paced, dynamic environment.

Professional Experience
Company : Avi Software
Role : Python Developer
Tenure : June 2020 - Till Now
Major Projects Undertaken:

Project: Automated Quality Inspection System for Manufacturing

Objective
Develop an automated quality inspection system that uses computer
vision to detect defects in manufactured products in real-time. The
system should identify and classify defects such as cracks, scratches, or
irregularities on various products.
Components of the Project
1. Data Collection and Annotation
o Gather Data: Collect high-resolution images of products
from the manufacturing line. Ensure you have a diverse
dataset representing different product types, defect types,
and manufacturing conditions.
o Annotate Data: Label the images with the types and
locations of defects. Tools like Labelbox, VGG Image
Annotator (VIA), or RectLabel can be used for annotation.
2. Preprocessing
o Image Preprocessing: Apply techniques like normalization,
resizing, and augmentation to prepare the data for training.
Data augmentation can include rotations, translations, and
color adjustments to improve model robustness.
o Segmentation: Segment the images to focus on areas of
interest, such as the surface of the product.
3. Model Development
o Object Detection: Use models like YOLO (You Only Look
Once), Faster R-CNN, or SSD (Single Shot MultiBox Detector)
to detect and classify defects.
 o Image Classification: Use Convolutional Neural Networks
(CNNs) such as ResNet, EfficientNet, or VGG for classification
tasks if detecting defects in different categories.
o Anomaly Detection: For detecting novel or unseen defects,
you could implement unsupervised anomaly detection
methods using autoencoders or generative adversarial
networks (GANs).
4. Integration and Real-Time Processing
o Real-Time Inference: Implement real-time image processing
using frameworks like OpenCV or TensorFlow Lite for edge
devices.
o System Integration: Integrate the computer vision model
with the manufacturing line's control system to flag
defective products for removal or further inspection.
5. User Interface
o Dashboard: Develop a dashboard to visualize inspection
results, defect statistics, and real-time alerts. Use tools like
Flask/Django for web applications or integrate with
industrial control systems.
6. Deployment
o Cloud Deployment: Alternatively, deploy the model on a
cloud platform (like AWS, Azure) for batch processing and
large-scale data analysis.
7. Testing and Evaluation
o Accuracy: Evaluate the model using metrics like Precision,
Recall, F1-Score, and Intersection over Union (IoU) for object
detection.
o Latency: Measure the inference time to ensure it meets real-
time processing requirements.
o User Feedback: Collect feedback from users to refine the
system and improve defect detection capabilities.
 8. Maintenance and Monitoring
o Model Retraining: Periodically retrain the model with new
data to adapt to changes in manufacturing processes or
defect types.
o System Monitoring: Implement monitoring tools to track
system performance, detect failures, and ensure continuous
operation.
Technologies and Tools
 Machine Learning Frameworks: TensorFlow, PyTorch, Keras

 Object Detection Models: YOLO, Faster R-CNN, SSD

 Image Processing Libraries: OpenCV, scikit-image

 Deployment: TensorFlow Lite, ONNX, Docker

 Web Frameworks: Django

 Cloud Platforms: AWS SageMaker, Azure Machine Learning

● Enhancing User Experience Using Large Language Models

○ Project Description
■ Tech companies often possess extensive collections of
helpbooks and support documents. However,
maintaining a dedicated blog website for these
resources is typically seen as an unnecessary overhead
that diverts attention from their core objectives.
Managing these documents can also be challenging,
frequently leading to duplication when documents
are requested.

■ This project aimed to create a sophisticated

platform that allows customers to upload their
documents and obtain insights through effective
indexing. Additionally, a state-of-the-art
recommender system was developed integrating
third-party data providers to suggest relevant
content based on user journeys. To enhance the user
experience, an interactive chat interface was
introduced, replacing cumbersome large documents.
This interface suggests questions and topics of
interest while tracking user engagement and
interactions. Essentially, the project integrates
advanced chatbot capabilities with recommender
systems to provide a more dynamic and user-friendly
support experience.

○ Tools and Technologies

Python, TensorFlow, PyTorch, ElasticSearch, FastAPI,
Docker, Kubernetes, AWS SageMaker, AWS Lambda, Redis,
LLMs, Weighted Word Embedding Models, Sentence
Embedding Models, LangChain, Haystack, Elasticsearch,
Transformers, Scikit-learn, 6sense, Clearbit, ZoomInfo,
Crunchbase, DataFox

○ Roles and Responsibilities

■ Led the creation of a comprehensive platform for
document indexing and insight generation using
cutting-edge NLP models. Utilized named entity
recognition, BERT and Roberta for understanding
and processing textual data to deliver insights such
assuggesting tags for an article and taxonomy
generation. Employed tools like NetworkX for network
analysis and Gensim for topic modelling to uncover
deeper insights. Implemented semantic search using
Elasticsearch using weighted word and sentence
embedding models to improve search relevancy.
■ Developed a sophisticated recommender system
leveraging collaborative filtering, and content-based
filtering techniques and integrating various data
sources such as 6sense, Clearbit, ZoomInfo,
Crunchbase, and DataFox to enrich user data and
enhance the accuracy and personalization of
recommendations. Utilized these sources to gather
insights about user preferences, behaviours, and
trends. to suggest relevant content based on user
journeys. Integrated models like Sentence-BERT as re-
rankers to improve recommendation accuracy.

■ Designed and implemented an interactive chat

interface using various proprietary and open-source
LLMs to enhance user engagement. This interface
dynamically suggests questions and topics of interest
based on real-time user interactions.

■ Deployed the solution on AWS SageMaker and

managed the infrastructure using Docker and
Kubernetes to ensure scalability and robustness.
Employed AWS Lambda and Redis for efficient task
management and caching.
Fraud detection
The model is being trained on a dataset that contains transactions between two
parties, first we process the data and perform feature engineering on it to reduce
the dimensions of the data. After we Train the model using decision trees algo,
the accuracy of the model is 82%.

Technologies used:
● python
● Pandas
 ● Sklearn
● AWS
Responsibilities:
● Processed the dataset using pandas
• Feature engineering using the concept of correlation
• Training the model using Decision Trees algo