Graduate student in "Scientific Computing" with comprehensive knowledge in data analysis and deep learning, high performance computing, numerical simulations. Extensive work experience as an automotive design engineer.
| Projects | Tech Stack | Brief description |
|---|---|---|
| Question_Answering_T5_BioASQ | transformers, Python (PyTorch, PyTorch lightning, Tensorboard, sklearn, NumPy, Pandas, and Matplotlib libraries) | BioASQ dataset was used to finetue and evaluate the performance of the pretrained t5-base model from huggingface transformers for question answering task |
| Sentiment Classifier using pretrained BERT model | transformers, google_play_scraper, Python (PyTorch, sklearn, NumPy, Pandas, Seaborn and Matplotlib libraries) | Reviews scraped from the Google Play top 10 apps in productivity category was categorized into Positive, Negative and Neutral reviews using pre trained BERT model from huggingface |
| Parallel Numerical Linear Algebra (PNLA) library for solving linear system of equations | c++, openMP, cmake, Doxygen, slurm | PCG solver for solving the Linear Equation System arising from different physical problems. CRS matrix storage and OpenMP was used. Effective scaling observed for a system of ~ 2.5 million unknowns when run on multiple threads |
| Image classification using resnet9 model | Python (PyTorch, NumPy, and Matplotlib libraries) | Used data normalization, data augmentation and regularization techniques to classify the images of the CIFAR-10 with an accuracy of ~ 91% in <5 mins |
| Similar document retrieval | Python (turicreate, sklearn, nltk, Pandas, NumPy, seaborn and Matplotlib libraries) | Clustering and information retrieval using nearest neighbor algorithm on a subset of wikipedia corpus |
| Analyzing product sentiment | Python (turicreate, sklearn, Pandas, nltk, NumPy, seaborn and Matplotlib libraries) | Classified the reviews based on bag-of-words(word count) and TF-IDF transformation with only top and bottom scoring words and compared the results with the complete superset of words |
| Personalized song recommendations | Python (turicreate, Pandas, sklearn, NumPy, and Matplotlib libraries) | Collaborative filtering and matrix factorization techniques were adapted to build an optimal and personalized song recommender |
| Simple Text generation using GPT-2 | transformers, PyTorch | Top-p sampling technique used to generate one or multiple texts of given sequcence length based on the input prompt text |
| Programming, Data Analysis, and Deep Learning in Python (Offered by "Serious Games Chair" by Prof. Dr. Jörg Müller WS 2020/21) | Python (Keras, sklearn, Pandas, NumPy, Matplotlib libraries) | Application of some interesting sub topics covered during the lectures and extension of some assignment problems |
| Linear Regression from scratch | Python (sklearn(for loading datasets), Pandas, NumPy, Matplotlib libraries) | Implemented all the functions required for model to perform both forward and backward propogation and evaluated the model on house price prediction of data from "Boston housing" dataset . Considered gradient descent algorithm for optimization and MSE as the loss function |
| Logistic_Regression_From_scratch | Python (sklearn(for loading datasets), Pandas, NumPy, Matplotlib libraries) | Implemented all the functions required for model to perform both forward and backward propogation and evaluated the model on multiclass classification of data from "Iris" dataset . Considered gradient descent algorithm for optimization and cross entropy as the loss function |