HowardMiller, USA

• Involved in AI strategy development with Deep Learning for optimizing budget distribution in PPC campaigns.
• Designed and developed new features, optimized existing ones, and integrated third-party API solutions.
• Tools: Tensorflow, Keras, Pandas, Scikit-Learn, Flask, Django, Celery, Redis, Postgres, Linux, Gunicorn, Apache, AWS, Reactjs, API, Docker

Gmonster, Hungary

• Oversaw the entire development pipeline.
• Designed and developed an entire product (Mass Email Marketing App) from scratch including the subscription system.
• Tools: Python, PyQt5, Database, Flask, Javascript, Linux Server

EnableGeek

• Designed and oversaw the entire operation pipeline.
• Managed both the Content Development team and the Tech team.

This research project, accomplished through a collaboration with a prominent professor in the United States, delves into the impacts of high-intensity light on object detection models. Our focus was primarily on exploring adversarial attacks with light and strategizing its defense mechanisms. We trained the YOLO5 model on a custom dataset, meticulously curated to encapsulate diverse lighting conditions. The model's performance under low light and nighttime scenarios was evaluated and further refined through increased representation of these conditions in the training dataset.

Our subsequent endeavors involved introducing light noise at calculated and random positions, the former derived from the gradient between the input and loss. The introduction of light noise had minimal impact on the model's performance, a finding that was substantiated through the application of a standard adversarial attack. We are currently in the process of documenting our research and findings in a journal article, where I will serve as the first author.

This project focused on creating a custom object detection and tracking system using Darknet-YOLO and Python. The first step involved assembling a unique dataset from videos taken by security cameras at fuel stations and highway traffic cameras. The data was then carefully annotated using YOLO mark, which set the stage for training the model.

The final phase involved integrating an Intersection over Union (IOU) tracker with the trained model to enable efficient object tracking. The project showcased an effective application of modern object detection and tracking techniques, offering a practical solution for real-world surveillance scenarios.

Developed a real-time speed detection system for vehicles using Darknet-yolo and Python. This system allows for instantaneous speed monitoring, aiding in traffic management and control.

This project involved creating an automatic vehicle entry management system for apartments. Using OCR and Darknet-yolo with Python, the system was developed to manage and monitor vehicle entries efficiently.

This project involved creating a custom dataset for training a YOLO (You Only Look Once) model. Different type of surveillance video was gathered from various sources. Tools utilized in this project included Yolo Mark and Linux Shell Scripting.

This experience dates back to my first year of undergraduate studies. After successfully passing an interview, I joined the team tasked with developing a unique chassis using PVC pipe, resulting in a basic cubic shape. We utilized Brushless DC (BLDC) motors to generate underwater thrust. Initially, we employed two motors for submersion and two for differential drive underwater.

We encountered some challenges in sinking the structure. To overcome this, we added two water-filled bottles, which allowed us to slightly sink the structure. However, a more effective solution was needed. We then decided to fill the structure itself with water and incorporated submersible pumps. This new arrangement enabled us to control the buoyancy of the structure by letting water in and out as needed.

Up until that point, we had been using wires for communication. Unfortunately, we realized that the signal strength weakened significantly with the length of the wire due to attenuation. This led our team to divide into two groups. One focused on solving the communication problem. This group later developed a "Low-Cost Underwater Wireless Communication System using Piezo-Ceramic Transducers", and their work was published in a research paper.

This project was centered around the RFID chip in our university ID cards, aiming to simplify attendance procedures. We used Python and Socket Programming for real-time data communication, and SQL for data storage. The outcome was a straightforward, automated attendance system that made good use of our existing resources.

This project aimed to create a Human Assistant - an autonomous device capable of navigating within a building and responding to calls to deliver light packages. The robot was built using a plastic water container, mounted on a rectangular steel chassis equipped with four DC motors. To navigate, it used a combination of Lidar for mapping the surroundings and sonar sensors for obstacle detection.

We initially considered using GPS for indoor navigation, but due to performance limitations, we devised an alternative approach. We placed unique markers throughout the building's interior and used a camera for image capture. The robot was programmed to recognize these markers and calculate its position and trajectory accordingly. The technical components of this project included Python, Raspberry Pi, Socket Programming, Image Processing, and Lidar. This innovative blend of technologies enabled us to accomplish our goal in a straightforward yet effective manner.

• Line Follower Robot with Automatic Collision Avoidance System
• Maze & Grid Solver Robot
• Remote Controlled Vehicle
• Auto Balancing Bot (Control System Project)
• DX Ball Game With Dot Matrix Display and Arduino (Microprocessor Project)
• Egg Incubator

• Tradeshow Lead Scanning Website
• Personal Trainer Website (Video)
• Lead Mining Website
• Smart Class Updates
• Scraping Application: Craiglist Scraper, Linkedin Scraper
• Automation Bot: Mass Email Marketing App(MVC), Auto SMS Sending Bot, Copy Assister App
• Desktop Application: Pi Camera Booth Controller with Python, Workout Generator for GYM (Video)