• Graded and provided feedback on 10+ assignments, ensuring consistent application of grading rubrics and enhancing student understanding, leading to a 15% improvement in class performance on assessments.• Guided 40+ students during office hours by offering personalized assistance on complex concepts, resulting in a 20% increase in student engagement and comprehension.• Responded to student queries within 24 hours, providing clear explanations and support, which improved student satisfaction rates in feedback surveys by 10%.• Assisted in the development of course materials and supported the professor in maintaining accurate academic records, contributing to 90% on-time submission of grades and assignments.

• Collaborated with a team of four to develop and document a functional AI-based Intrusion Detection System (IDS), meeting all milestones and contributing valuable insights to the cybersecurity research community.• Architected a custom SecBERT-based one-class classification model specifically for network anomaly detection, achieving a 15% loss reduction after early stopping, optimizing threat detection in a large-scale network environment.• Developed efficient Python scripts that processed up to 124 million rows of data for real-time IP address comparison, enhancing the system’s ability to detect and respond to threats within milliseconds. This significantly improved the efficiency and reliability of network traffic analysis.• Two-Tier Anomaly Detection Approaches: • Approach 1: SecBERT One-Class Classification: Architected and fine-tuned a one-class classification model using SecBERT, modifying the architecture by incorporating a sigmoid activation function and Binary Cross-Entropy (BCE) as the loss function. This approach resulted in a 15% reduction in detection loss after implementing early stopping, enhancing the model’s capability to accurately identify network anomalies. • Approach 2: SecBERT with Isolation Forest: Integrated SecBERT embeddings with an Isolation Forest model, achieving 79% accuracy in detecting outliers within an unsupervised one-class scenario. This approach significantly improved the IDS’s ability to identify sophisticated cyber threats.• Managed and processed over 130,000+ network traffic records using advanced preprocessing techniques, including normalization and feature extraction, to prepare data for model training. Overcame the challenge of unlabeled data, enabling effective unsupervised learning.• Utilized Plotly to generate critical visualizations that revealed key relationships between packet length and protocol type, which were essential for refining the model's behavior and improving overall detection accuracy.

• Completed a simulation focused on advising a client on customer targeting with the Data, Analytics & Modelling team• Assessed data quality and completeness in preparation for analysis• Analysed 20,000 data transactions to target high-value customers based on demographics and attributes• Developed dashboard in Microsoft Power Bi to communicate findings with visuals

• Collaborated within a 9-member team, serving as the project scribe responsible for data visualization. • Analysed key performance indicators for 11 campaigns within Facebook's SuperHeroU campaign dataset. Evaluated KPI metrics like average spend, CPR, CPC, Reach, and Impressions. • Identified potential for a 14.5% income increase by excluding campaigns 3 and 10. • Segregated predominant reach from the 18-24 years age group, followed by the 13-17 years age group.

• Monitored and optimized AWS resources (S3, RDS) to ensure efficient relational database data storage and retrieval, improving data availability and reliability for the Alphabet News app.• Reduced storage costs by 15% by transitioning archival data to S3 Glacier, optimizing cloud storage for large-scale data handling.• Generated scalable data pipelines using AWS Lambda and Kinesis to automate real-time processing of 40+ daily news articles, enabling timely data integration and analysis.• Optimized data workflows using distributed computing (Kinesis and Lambda), significantly reducing data latency and improving real-time analytics performance.• Orchestrated ETL pipelines using AWS Glue and Step Functions, minimizing manual intervention and cutting data processing time by 40%.

• Worked and developed POC’s for industries seeking deep learning solutions. Involved with the CEO for future development and was an active member of the Tech Team.• Built and optimized TinyML models for deep learning applications in Edge A.I. world.• Performed experiments on Edge Impulse platform to collect the data required and to build deep learning model for Air Gesture classifier of 0-9 digits with 87% accuracy and deployed it on the development board for inferencing.• Developed web-based dashboards using Node-Red and Streamlit (Python Library) to analyze the data displayed on them from 5-6 sensors.• Applied Deep Learning and Machine Learning models based on the applications of 3 v’s which were used in industry 4.0 (Vision, Vibration, Voice).

• Gained familiarity with various acoustic and motion sensors.• Learned fundamental basics about microcontroller (Arduino Uno).• Worked as part of a team on a Wheelchair Controller project.• Utilized BLDC Motor, PWM generator, and Arduino Uno for the project.• Developed a system to enhance the maneuverability of the wheelchair.• Aimed to create a solution that benefits people in need by making wheelchair movement easier.