For many people today, using speech to communicate with computers is a common task. Intelligent personal assistants (IPAs) such as Amazon Alexa, Apple Siri, Microsoft Cortana, or Google Assistant allow people to perform various tasks such as searching, scheduling a meeting, or making a hands-free phone call without having to hold a mobile device or interact with a graphical user interface. These intelligent assistants use natural language user interfaces (NLUIs) to interact with users. However, these assistants are currently unable to perform tasks that are possible using transformer-based large language models (LLMs). The revolutionary LLMs are capable of advanced reasoning (as in GPT-4) in addition to information retrieval. To harness this potential, I used Python libraries such as OpenAI's general-purpose speech recognition model Whisper together with Amazon Polly's API to create an IPA capable of voice activation, speech transcription in different languages, and natural-sounding human speech synthesis based on Amazon Polly's neural-based models.

Today, machine learning and deep learning approaches are widely used to analyze the vast amount of data collected from the Internet. However, manual data collection is both error-prone and time-consuming, especially when dealing with unstructured or semi-structured data. Web scraping is commonly used to address the labor-intensive and error-prone process of manual data collection. In this work, I explored methods to develop a state-of-the-art system for locating, collecting, and classifying biodiversity funding opportunities in Germany. Using the spiders I developed, I assembled a dataset and trained classifiers that were able to successfully achieve the desired classification performance, according to the defined metrics.

It is known that more than 75% of all day traders end up quitting within just two years. According to a study by multi-asset online trading platform eToro, 80% of day traders lost money over the course of a year, with the median loss being 36.30%. Algorithmic trading is a method of executing orders using automated, pre-programmed trading instructions that take into account various influential market variables. Algorithmic trading takes advantage of the superior speed and data processing capabilities of computers over human traders. Are price movements in the cryptocurrency market predictable? In this work, I analyzed various traditional time series analysis methods, such as ARIMA and seasonal ARIMA, as well as classification and regression machine learning algorithms to identify patterns and extract insights from the cryptocurrency market data.

I developed an application using the Vuforia and Unity engines to augment the figures and graphs presented in the paper with additional data, such as visual elements. The idea is to help the reader gain a better understanding of the concepts presented in the graphs, tables, or figures by providing additional visual cues. I used the C# programming language to write the application logic. A simple and intuitive GUI was used to navigate between menus along with some basic functionality such as hiding/showing the augmented text description. I also added video, sound, and animation to the visual elements in the scene to make a more elaborate experience.

I was able to easily perform denial of service, eavesdropping, and hijacking attacks on a Bluetooth-enabled light bulb. To perform the attacks, I used some of the tools available on Santoku and Kali Linux. In addition, I used a commercial off-the-shelf microcontroller, originally designed for educational use in the UK schools, to perform the attacks. Furthermore, I reverse-engineered the manufacturer's application for Android to create an application using JavaScript, HTML, and CSS. Consequently, I took control of the device and was able to replicate much of the functionality of the original application. The results obtained demonstrated the vulnerabilities of Bluetooth Low Energy 4.1.

BLE 4.0 and 4.1 have been shown to be vulnerable to passive eavesdropping attacks. However, in many cases, the necessary security mechanisms are not implemented in devices equipped with these versions of BLE. Pairing is a mechanism that provides encryption for BLE communications using the Just Works, Passkey Entry, and Out of Bound pairing schemes. However, the first two pairing schemes do not provide eavesdropping protection. Application-level authentication could be implemented to prevent eavesdropping and unauthorized access. Bluetooth version 4.2 introduces LE Secure Connections, which incorporates the Elliptic Curve Diffie-Hellman key agreement protocol. This mechanism makes BLE 4.2 and later versions more resistant to passive eavesdropping.

Traditionally, existing streaming camera architectures use power-hungry and compute-intensive components such as analog-to-digital converters and video codec compressors. In the paper "Towards Battery-Free HD Video Streaming," a group of researchers from the University of Washington focus on a novel approach to streaming high-definition (HD) video using a novel analog video backscatter scheme and two complementary techniques as enablers. They design a high-definition video streaming system using a low-power architecture. In principle, this architecture allows the analog values associated with each photodiode in the image sensor of portable or fixed cameras to be processed and sent directly to the reader. As a result, power-intensive tasks are delegated from the camera to the reader. This reduces the energy required for video streaming by several orders of magnitude.

Automatic Dependent Surveillance-Broadcast (ADS-B) is the next generation of air traffic control technology that the Federal Aviation Administration (FAA) and the European Aviation Safety Agency (EASA) require the vast majority of aircraft to be equipped with by January and June 2020, respectively. However, ADS-B as it is used today is not without its problems. This paper discusses observed attacks and their methodologies that can be realized over ADS-B surveillance technology and compromise its availability, integrity, or confidentiality. As an introduction, the paper examines the basic characteristics of ADS-B, functionality, deployment status, and other relevant facts and figures to better understand the fundamentals, current state, and pitfalls of the state-of-the-art technology for air traffic surveillance and control. The paper then examines the complexity, severity, and impact of the attacks by presenting some examples. Finally, the paper complements the previous sections by providing an overview of proposed solutions and mitigation techniques, and reviews the most prominent ones.

For my undergraduate thesis, I designed and developed a device to measure vital signs (pulse rate and body temperature). I recorded pulse rate using a reflection-type photoplethysmography (PPG) pulse sensor from the user's fingertip and used a digital temperature sensor to measure auxiliary body temperature. The readings were first displayed on a screen. In addition, I was able to retrieve them via cellular networks using SMS communication. Moreover, I developed a web service that allowed the device to connect to the web server via an API endpoint and submit data for processing. Finally, I designed a graphical user interface to display the extracted insights to the user. I wrote the embedded software in C and C++ and designed the PCB using Altium Designer. The heart of the device was an Atmel ATMega32 microcontroller. To add telemetry and remote control capabilities to the device, I integrated a SIM900A GSM/GPRS module into the circuit.

After launching my career with a graphic design service website, I ventured into the e-commerce world, selling a diverse range of products from plush toys to wristwatches. Initially, I managed all aspects of the business, including development, operations, and sales. This encompassed website design and administration, marketing, logistics, and customer service—except for shipping, cash on delivery, and payment processing. During the initial phase as an online merchant, I processed over 1,200 sales orders through one of my stores and achieved a total revenue of over 230,000,000 IRR. Later, I delegated the logistics, customer support, and order processing through affiliate marketing and was rewarded as the most profitable affiliate partner by a popular affiliate marketing network at that time, called Bermoda (figures not included).