Abhijeet Bajaj Research Profile

About

Abhijeet Bajaj is a skilled software engineer with expertise in machine learning, distributed systems, and large-scale software development. With a Master’s degree in Computer Science from Columbia University (CGPA: 3.83/4.0) and a Bachelor of Engineering in Computer Science from BITS Pilani (CGPA: 8.43/10.0), he possesses a strong academic foundation in algorithms, databases, and AI-driven applications. Currently, he works as a Software Engineer II at Uber, where he has played a crucial role in enhancing the marketplace dynamics through driver surge pricing algorithms. His work has contributed to launching novel pricing models in 15 Latin American cities and building metric pipelines to monitor the health of these systems. Prior to Uber, Abhijeet gained valuable experience as a Software Engineer Intern at Google Cloud Network, where he developed a real-time dashboard to monitor network health and implemented a command-line interface (CLI) that detects and analyzes anomalous events. His work involved utilizing DBSCAN clustering techniques to find correlations in network failures, significantly reducing on-call troubleshooting time. Additionally, he worked as a Strategy Analyst at Goldman Sachs, where he was responsible for reviewing bonds for the Rates & Credits team and monitoring software-based control mechanisms in electronic trading. His experience at Samsung Research as a Research Intern further honed his expertise in computer vision, where he worked on depth-based annotation techniques for RGB-Depth images and improved upon methodologies for real-time human pose estimation. His technical proficiency spans multiple programming languages, including C/C++, Python, and Java, as well as deep learning frameworks such as PyTorch, TensorFlow, and Keras. His coursework in distributed systems, data structures, and object-oriented programming has provided him with the knowledge to solve complex engineering problems. Throughout his career, Abhijeet has demonstrated a strong aptitude for problem-solving and optimization, evident in his projects like a multi-threaded chatbot with multimodal input and an A* search-based pathfinding solution for Google Maps covering 40 square kilometers of New Delhi. His approach to software development integrates AI-driven solutions with practical applications, as seen in his contributions to anomaly detection, trading risk mitigation, and AI-based perception models. He is also a winner of the 2019 Tesco Graduate Hackathon, where he co-developed a novel data lineage extractor that visualized SQL script dependencies through a collapsible tree. His commitment to innovation, along with his ability to manage concurrent processes, optimize data structures, and design scalable AI models, makes him a highly valuable asset in the software engineering landscape. With expertise in REST APIs, Git, JIRA, SQL, and various AI and ML libraries, Abhijeet continues to drive impactful projects that bridge the gap between theoretical advancements and real-world applications.

Skills

Experience

Software Engineer 2

Uber, USA

Feb-2022 to Present

Education

Columbia University (CU)

M.SC in Computer Science

Passout Year: 2021

Publication

Research Article

Abhijeet Bajaj

December, 2024

Predictive Surge Pricing Model for On-Demand Services Based on Real-Time Data

Journal : IJRMEET - International Journal of Research in Modern Engineering & Emerging Technology

In the evolving landscape of on-demand services, surge pricing has emerged as a critical pricing strategy to balance supply and demand in real-time. This research explores the development of...

Peer-Reviewed Articles

COMPARATIVE ANALYSIS OF REVERSE IMAGE SEARCH ENGINES USING DIVERSE IMAGE SETS

Eight well-known reverse image search engines—Google, Bing, TinEye, Yandex, Baidu, Getty Images, Shutterstock, and Alamy—are compared in this study based on several different factors. Language support, speed, accuracy, facial recognition, geographic coverage, cropping feature, number of images retrieved, ease of use, mobile app availability, privacy measures, input options, supported file formats, search methods, and additional features are some of these requirements. The study outlines each engine's advantages and disadvantages. Both Google and Bing are very user-friendly, fast, and support multiple languages. However, Google is more accurate and has features like facial recognition and SafeSearch. Yandex offers comparable functionality but targets the Russian market. TinEye promotes privacy and collects very little data, however, it has trouble with unique photos and doesn't have many sophisticated capabilities. Baidu offers little privacy and openness and caters mostly to the Chinese market. Although Shutterstock and Getty Images have extensive privacy policies, their accuracy is not as high. Alamy has a reduced precision of retrieval but complies with data standards. According to the analysis, each engine serves a particular purpose. Google or Bing may be preferred by users looking for smart image detection and user-friendliness. TinEye might work for users who are concerned about their privacy. In the end, the decision is based on personal preferences and search objectives.

DESIGN AND IMPLEMENTATION OF Wi-Fi DEAUTHENTICATION SYSTEM USING NODEMCU ESP8266

Network security is seriously threatened by Wi-Fi de-authentication attacks, which frequently lead to data interception, illegal access, and service interruption. The mechanics and ramifications of these assaults are explored in detail in this research study, which highlights how they could jeopardize network availability, secrecy, and integrity. To bridge theoretical understanding with actual experimentation, the paper presents a practical implementation of a Wi-Fi deauther utilizing the NodeMCU ESP8266 microcontroller platform. With the use of programs like the Arduino IDE and NodeMCU Flasher, the Wi-Fi deauther was created and put through testing to identify and stop de-authentication threats instantly. The system's high detection accuracy, quick response times, and little effect on network performance as a whole are demonstrated by the experimental findings. The NodeMCU ESP8266 platform demonstrated good resource management by managing the detection and countermeasures while keeping CPU use below 70% and guaranteeing less than 5% reduction in network performance and latency. This study advances wireless network security by demonstrating a scalable, affordable method of thwarting de-authentication attacks and by suggesting further improvements that would include machine learning integration and wider assault coverage. For network managers, cybersecurity experts, and researchers looking to strengthen wireless network defenses, the findings offer insightful information and useful recommendations.

Blockchain using Virtual TRY-ON

In today’s dynamic retail environment, the shift towards online shopping necessitates innovative solutions that enhance customer engagement and satisfaction. This project introduces a virtual try-on clothing platform designed to revolutionize the online shopping experience by merging cutting-edge augmented reality (AR) and machine learning technologies. The platform enables users to visualize how garments will fit and appear on their unique body shapes without the need to visit a physical store. By offering a user-friendly interface, the website allows customers to upload personal images or utilize real-time video features, facilitating an interactive and personalized shopping experience. Key functionalities include accurate size recommendations tailored to individual measurements, as well as curated fashion suggestions that align with users' personal styles. These enhancements aim to minimize return rates—a significant challenge in e-commerce—while simultaneously boosting customer satisfaction and driving sales. Additionally, the platform fosters social interaction through built-in sharing capabilities, allowing users to solicit feedback from friends and family, thus enriching the decision-making process. This aspect not only enhances the shopping experience but also builds a sense of community around fashion choices. By integrating advanced technology with a seamless and engaging user experience, this virtual try-on website represents a substantial advancement in online fashion retail. It sets the stage for a more personalized and interactive shopping journey, ultimately redefining how consumers engage with fashion in the digital age. As we look to the future, this platform aims to become a cornerstone of online retail, reflecting the evolving needs and preferences of today’s consumers.

The Power of AI and Machine Learning in Cybersecurity: Innovations and Challenges

Networks and sensitive data are no longer adequately protected by traditional security methods due to the ongoing evolution and sophistication of cyber-attacks. Cybersecurity can be enhanced with the exploitation of machine learning and artificial intelligence techniques, which make threat detection more effective and efficient. This article, while giving an outline of the field's present position, discusses the difficulties in adapting machine learning and artificial intelligence (ML) to cybersecurity. The research discusses machine learning methods that are applied to tasks like malware classification, anomaly detection, and network intrusion detection. Lastly, the necessity for sizable labeled datasets, the adversarial attacks on machine learning models, and the adversity of deciphering models of black-box ML are some of the boundaries and challenges that are also covered.

Design of 4-bit ALU for low-power and High-speed Applications.

This paper presents a novel design and optimization of a 4-bit Arithmetic Logic Unit (ALU) utilizing 90nm CMOS technology, specifically addressing the longstanding carry-out issue prevalent in existing architectures. Notably, our proposed 4-bit ALU architecture successfully minimizes delay and power consumption by incorporating an optimized carry-out design employing AND gates. A comprehensive comparison of three logic styles - Pass Transistor Logic (PTL), Complementary Metal-Oxide-Semiconductor (CMOS), and Transmission Gate Logic (TGL) - is conducted, yielding significant improvements in power-delay tradeoffs. Simulation results validate the efficacy of our design in resolving the carry-out issue, making it an attractive solution for low-power, high-speed digital applications.

Deep Learning for Polymer Classification: Automating Categorization of Peptides, Plastics, and Oligosaccharides

Polymers represent a diverse and vital class of materials across numerous industries, each with unique structural characteristics and functional properties. Traditional methods of polymer classification rely heavily on labor-intensive techniques prone to subjectivity and human error. The emergence of deep learning has significantly transformed material science by enabling automated analysis and classification of complex polymers. In this study, we focus on leveraging deep learning models to classify three distinct classes of polymers: peptides, plastics, and oligosaccharides. Peptides, plastics, and oligosaccharides represent significant subsets of the polymer family, each with distinct structural features and applications. Our research explores the effectiveness of various deep learning architectures, including deep learning to classify peptides, plastics, and oligosaccharides, achieving perfect accuracy with neural networks, K-Nearest Neighbors, and Random Forest classifiers. Principal Component Analysis enabled visualization of sample distribution, demonstrating deep learning's potential to automate and enhance polymer classification, reducing reliance on traditional, labor-intensive methods.

Detecting Fake Reviews in E-Commerce: A Deep Learning-Based Review

E-commerce platforms are increasingly vulnerable to fake reviews, which can distort product ratings and mislead consumers. Detecting these fraudulent reviews is critical to maintaining trust and transparency in online marketplaces. This review provides a comprehensive analysis of deep learning techniques used for fake review detection. Key models such as Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM) networks, and Transformer-based models like BERT are explored for their ability to analyze textual data and detect linguistic anomalies. Additionally, behavioral analysis using Convolutional Neural Networks (CNNs) and hybrid models combining textual and behavioral features are discussed. The review also highlights the role of Graph Neural Networks (GNNs) for network analysis and unsupervised learning methods like autoencoders for anomaly detection. Despite advances, challenges such as evolving fake review tactics, data imbalance, and cross-platform adaptability remain. The paper concludes by discussing future research directions, including enhancing model interpretability and combining deep learning with blockchain for more secure and verified review systems.

Harnessing Deep Learning for Precision Cotton Disease Detection: A Comprehensive Review

Cotton cultivation plays a critical role in global agriculture, yet its productivity is significantly hindered by various plant diseases that impact yield and quality. Conventional disease detection methods often fall short due to their reliance on manual inspection and limited accuracy. This comprehensive review explores the application of deep learning techniques beyond Convolutional Neural Networks (CNNs) in enhancing cotton disease detection. The paper covers a range of deep learning methodologies, including CNNs, Recurrent Neural Networks (RNNs), and hybrid models that combine different neural network architectures. It examines how these techniques can improve the precision and efficiency of disease diagnosis for common cotton ailments such as boll rot, leaf spot, cotton wilt, and bacterial blight. By reviewing current research and case studies, the paper provides insights into the effectiveness of various deep learning approaches and their integration into practical agricultural systems. It also addresses the challenges faced in implementing these technologies and suggests future directions for advancing disease management strategies through deep learning. This review aims to offer a holistic perspective on the potential of deep learning to transform cotton disease detection and contribute to more sustainable agricultural practices.

Face Recognition : Diversified

This paper presents a novel lightweight hybrid architecture for face recognition, combining the strengths of MobileNet and attention mechanisms to enhance performance under challenging conditions such as facial occlusions (e.g., masks), varied illumination, and diverse expressions. The proposed model is evaluated against popular baseline models, including MobileNetV2, EfficientNetB2, and VGG16, on the Yale Face Dataset and a Simulated Masked Yale Dataset. On the Yale Dataset, the hybrid model achieved superior results with an accuracy of 93.78%, precision of 94.45%, recall of 93.33%, and F1-score of 93.89%, outperforming the baseline models in all key metrics. Additionally, when tested on the Simulated Masked Yale Dataset, the hybrid model exhibited increased resilience to occlusion with an accuracy of 63.45% and F1-score of 64.22%, significantly surpassing the other architectures.

Blockchain in Cybersecurity: Enhancing Trust and Resilience in the Digital Age

Blockchain technology has emerged as a transformative tool in the field of cybersecurity, offering a decentralized, immutable, and transparent framework to enhance trust and resilience in digital systems. This review explores the various applications of blockchain in cybersecurity, focusing on its ability to mitigate key security challenges such as data tampering, unauthorized access, and identity fraud. By analyzing the integration of blockchain in areas like secure data sharing, IoT security, and identity management, this paper highlights the strengths and limitations of blockchain-based solutions. Furthermore, it examines consensus mechanisms and cryptographic techniques that ensure the integrity and confidentiality of information. Despite its potential, blockchain faces challenges such as scalability, regulatory hurdles, and susceptibility to attacks like 51% and Sybil attacks. This review aims to provide a comprehensive understanding of blockchain's role in enhancing cybersecurity, while also identifying future research directions to overcome current limitations.