Pulkit Arora Email & Phone Number

Rotterdam, Zh, Nl

Assen, Drenthe, Netherlands

Aachen, North Rhine-Westphalia, Germany

• Evaluated robustness of LLMs with adversarial paraphrasing. Conducting experiments on LLaMA2 on linguistic perturbations that provide an understanding of LLM’s ability to handle real-world language scenarios in QA tasks• Performed prompt-engineering to obtain consistent and reliable paraphrases from LLM output• Finetuned TempoBERT on time-separated English text to create a temporal text classification model• Used statistical tests, explainable AI methods, and semantic filtering to gain insights about model robustness

Erlangen, Bavaria, Germany

• Developed training and monitoring pipelines for PyTorch models on GPU clusters using kubeflow workflows• Conducted research on self-supervised representation learning using multimodal Transformer for emotion recognition• Implemented supervised and unsupervised transformer models for (neutral, stress, amusement) state prediction using ECG/EDA bio-signals. Achieved an accuracy of 98.8% with the WESAD dataset• Presented findings at ACII 2023 MIT Media Lab, Cambridge, USA

• Developed a knowledge graph using cutting-edge NLP algorithms to create hyperlinks in technical documentation• Finetuned BERT-based models to identify key concepts in documents• Used AI-based SOTA methods for key phrase identification combined with heuristics to rank the relevantkey-phrases in a document• Optimized the system to reduce document processing time by 80% through distributed computing and caching techniques.• Implemented the pipeline on AWS to reduce setup time for training and dataset uploads.

New Delhi, Delhi, India

• Developed and deployed machine learning models using Flask API in Docker environment.• Trained ML models for classifying user-company tweets based on complaint type and sentiment intensity using various approaches such as tfidf+MultinomialNB, word2vec, SVM, and XGBoost.• Developed an OpenCV-based solution for AR ring augmentation and hand segmentation in Android application

Gurgaon, India

• Developed vision based quality checking methods to ensure high-quality production standards• Designed a multi-threaded application using PyQt5 to perform real-time quality checking on camera streams• Configured a server to deploy quality checking tasks on NVIDIA Jetson Nano for efficient production line monitoring

M.Sc. Computer Science (Media Informatics)

Bachelor Of Technology, Computer Science