Hi, I’m Adnan Rizvi, a Computer Science Engineering graduate from VIT Vellore,
with a CGPA of 8.6.I have practical experience in data engineering, machine learning,
data visualization, and web scraping.
I’ve also worked on impactful projects like crime classification using deep learning,
spam SMS detection, and credit card fraud detection—each focused on delivering
high accuracy and real-world applicability.
I’m also proficient in web scraping using Scrapy, handling everything from crawling
and data extraction to managing logins, pagination, and data export.
My technical toolkit includes Python, Java, SQL, TensorFlow, AWS, and Tableau.
Beyond tech, I enjoy solving real-world challenges through coding, exploring
emerging technologies, watching movies, traveling, and spending quality time with
my family.
✅ Project Overview:
“I built a Scientific Calculator API using Flask that simulates how data pipelines
ingest, process, and return structured data. It supports both JSON payloads and query
parameters, making it flexible for various client inputs.”
✅ Key Features:
● Core Functionality:
The API handles basic operations like addition, subtraction, multiplication,
and division, as well as scientific functions like sqrt, log, sin, cos, and tan.
● REST Endpoints:
I implemented three main routes:
○ POST /calculate – Accepts JSON input for operation and operands.
○ GET /calculate – Accepts query parameters (operation, a, b) to perform
calculations.
○ GET /history – Returns all past calculations.
○ DELETE /history – Clears the calculation history.
● Modular Design:
I separated the business logic into a different file ([Link]) to keep the
code clean and maintainable.
● Postman Testing:
I used Postman to simulate API calls with both JSON and query parameters,
testing various edge cases like division by zero or missing parameters.
✅ Data Engineering Angle:
“This project helped me simulate a data processing pipeline where incoming data
(operands and operation type) are validated, processed through custom logic, and
returned as structured responses (like transformed output or logs). The /history
endpoint works like a temporary store or cache, mimicking a mini log database.”
✅ Tech Stack:
● Python, Flask (for API)
● Postman (for testing)
● Modular Python scripting to simulate scalable logic
✅ Why This Project Matters:
"This project sharpened my skills in building APIs, handling
structured/unstructured inputs, and simulating end-to-end
processing—skills directly aligned with real-world data engineering
pipelines.”
#####################################################################
###############
Project Title:
Crime classification using Artificial Intelligence with Deep Learning
🧠 What is the project about (in simple terms)?
I built a system that can watch surveillance videos and automatically
detect crimes like robbery, assault, vandalism, etc. It helps police or
security teams respond faster without manually checking hours of CCTV
footage.
🛠️ What technologies did I use?
Task Technology Used
Programming Language Python
Model Development TensorFlow, Keras (used DenseNet121)
Data Handling Pandas, NumPy
Preprocessing & OpenCV, ImageDataGenerator (Keras)
Augmentation
Frontend ReactJS, HTML, CSS, JavaScript
Backend API Flask (Python web framework)
Model Deployment Flask API + AWS Cloud
Evaluation Metrics ROC-AUC, Confusion Matrix, Accuracy,
F1-Score
Database Used to log predictions and user feedback
🔍 How does it work?
1. User uploads a video or image on the web interface.
2. The backend (Flask API) preprocesses it and sends it to the trained Deep
Learning model.
3. The model classifies it into one of 14 crime categories (like shooting,
shoplifting, arson, etc.).
4. The result is shown on the screen, and the user can also report the crime.
5. All reports and predictions are stored in a database, and feedback helps the
model improve over time.
Why DenseNet121?
Pre-trained on ImageNet: You used transfer learning by loading the pre-trained
DenseNet121 model.
Efficient Feature Learning: It connects each layer to every other layer in a
feed-forward manner, which helps in learning better features for complex images.
Faster and Accurate: DenseNet is known for reducing the number of parameters while
maintaining high accuracy.
🧠 In your project, DenseNet121 was used for:
Extracting features from crime scene video frames.
Classifying the input into one of 14 crime types (e.g., robbery, assault, vandalism,
etc.).
Fine-tuned on the UCF Crime Dataset (you froze initial layers and trained the rest).
📊 Results
● ROC-AUC Score: 0.84
● Very high accuracy for crimes like robbery, assault, and abuse
● Supports real-time classification and can be connected to live CCTV feeds
🌐 Why is this useful?
This system helps cities become smarter and safer by automating video
surveillance. It can reduce the workload of humans and improve crime
response time — useful for police, government, and security systems.
###################
During my summer internship, I worked on a project titled 'Integrated Supply
Chain Management,' where I designed and implemented a data-driven vehicle
compliance tracking system.
The core objective was to automate the entry-level inspection of vehicles by
categorizing them into Red, Orange, or Green states based on specific functional
criteria—like headlight and side mirror status. I developed a structured data
pipeline that captured and logged inspection results into a month-wise organized
database.
The system leveraged automated checks to classify vehicle entries and maintain
real-time tracking. I also built visual dashboards to analyze trends in vehicle
status over time—both percentage-wise and vehicle-wise. This involved
generating analytics on compliance rates and actionable insights for maintenance
planning.
Technically, my work focused on process automation, efficient data handling,
monthly database structuring (probably using Excel/Sheets or SQL), and
presenting results using visual analytics. This hands-on experience strengthened
my skills in data management, logic-based classification systems, and operational
reporting aligned with real-world industrial use cases."
"Let me walk you through the end-to-end workflow I implemented during my
internship project on vehicle compliance tracking under Integrated Supply
Chain Management.
1. Vehicle Entry Check:
Vehicles entering the designated area go through a step-wise inspection. Based
on issues like non-functional headlights or side mirrors, they are categorized as
Red, Orange, or Green.
2. Database Management:
I built a structured monthly database using Excel to log every vehicle's entry
and inspection status. This helped organize data chronologically and simplify
record-keeping.
3. Data Entry and Storage:
Each vehicle’s inspection result was entered into Excel, tagged with status
labels and other relevant identifiers, allowing us to maintain clean, analyzable
datasets.
4. Excel Data Processing:
I used Excel functions and formulas to preprocess data—calculate vehicle
counts, monthly trends, and prepare sheets for visualization. This included
aggregations and logical checks.
5. Integration with Tableau:
The Excel sheets were integrated into Tableau as a data source, allowing for
dynamic dashboard creation.
6. Data Visualization and Analysis:
I developed interactive dashboards and graphs in Tableau to show monthly
compliance rates, vehicle status breakdowns, and trends. This made insights
more accessible to stakeholders.
7. Reporting and Decision-Making:
Finally, these dashboards supported decision-making by highlighting key areas
of concern—like the number of non-compliant vehicles—helping the
management take proactive maintenance and safety actions."