Education

University of Massachusetts Amherst (Sep 2021 – Dec 2022)
M.S. Computer Science
GPA: 3.75/4.0

Shiv Nadar University (July 2016 – July 2020)
B.Tech. Computer Science and Engineering with minor in Mathematics
GPA: 9.1/10

Professional Experience

Roostify (June - Aug 2022)
Engineering Intern

• Automating a bank-transaction category identification task using transformer based models trained on the transaction description text. Framework: Pytorch, HuggingFace - Transformers
• Creating visualisation reports and dashboards using SQL and LookML for gaining insights into mortgage underwriting conditions data

Publications

In chronological order:

1. Mishra, P., & Mittal, R. (2021). NeuralNERE: Neural Named Entity Relationship Extraction for End-to-End Climate Change Knowledge Graph Construction. In Tackling Climate Change with Machine Learning Workshop at ICML.
2. Jain, S., Mittal, R., Jain, A., & Sinha, A. (2019, September). An efficient framework for monitoring tree cover in an area through aerial images. In Applications of machine learning (Vol. 11139, pp. 327-333). SPIE.
3. Jain, S., Mittal, R., Mishra, P., & Sinha, A. (2019, September). Road network mapping from aerial images. In Applications of Machine Learning (Vol. 11139, pp. 321-326). SPIE.

Projects

Climate change knowledge graph construction | Natural Language Processing

• Developed an end-to-end deep learning based model for extracting the intended relationship phrase given subject and object entity pairs directly using raw text of relevant news articles
• Proposed an unsupervised learning scheme for obtaining the subject - relationship - object triplets for constructing a knowledge graph
• Created the dataset by web scraping over 11k climate change news articles from the Science Daily website
• Accepted in Tackling Climate Change using Machine Learning workshop at ICML 2021

Sentiment Analysis on Code-Mixed Languages | Python, Pytorch, HuggingFace

• Fine-tuned a baseline Distil m-BERT model for the task of sentiment analysis on code-mixed Hinglish language
• Implemented data augmentation techniques including backtranslation using two separately trained mt5 models; as well as transliteration through Google translate
• Implemented a modified tokenization scheme by tuning the model tokenizer on the training set
• Visualized attention maps of the sentiment analysis models for increased model interpretability

Music Thumbnailing through Artist Recognition | Python, Pytorch, Scikit-Learn, NumPy

• Implemented and trained an attention-based neural network model on the task of music artist recognition
• Generated music thumbnails using attention scores for each song segment
• Verified the efficacy of the attention based model by evaluating the attention score results using a separately trained convolutional recurrent neural network (CRNN) model
• Observed an increase from 55% accuracy on the lowest attention score segments to 74% accuracy on the highest attention score segments using the CRNN model

Monitoring tree cover in an area through aerial images | Python, Tensorflow, NumPy

• Adopted the Mask R-CNN model for the task of tree instance segmentation and counting, through transfer learning
• Used a Gist Feature based sampling technique to minimize the amount of training data required for scaling the model to regions with varied geographical features
• Published in Proc. SPIE, Applications of Machine Learning 2019

Covid-19 Visualisation Dashboard | Javascript, D3.js, HTML, CSS, Bootstrap

• Built an interactive covid-19 visualization dashboard for depicting the number of covid cases, state-wise, through time using a color gradient based approach
• Combined the covid-19 data with state-wise mental health care data to interactively visualize the trend between covid-19 and mental health cases in each state