• Worked on improving functionality of existing NodeJS based APIs, which were being used in the company’s internal dashboard. I was also responsible for the front-end changes and writing of Unit Tests (UTs) for the APIs.
• Was also able to convert my internship into a full-time offer.
• Created custom OpenAPI schema validation rules. These custom rules allows much stricter schema validation.
• Created an automated python script to extract out 4 pipeline data (UT, Lint, Type Errors, Circular Dependencies) from each of the 25 microservices.
• Integrated In-Memory MongoDB with Jest, which eliminates the need to mock DB calls for UTs.

Tech stack: NodeJS, Typescript, OpenAPI, MongoDB, Jest, MySQL, Grafana, Kibana, Microservices and Distributed Systems.

• Explored and implemented topic modelling algorithms like Word2Vec, LDA, and Lda2Vec using Gensim and PyTorch in a given knowledge domain of water resources management. The experiments were conducted on 28k full-text documents and 88k abstracts over 7 academic journals.
• Generated word clouds and interactive HTML visualizations showing the inter-topic distance map using PyLDAVis.
• Computed and compared topics generated by each model [LDA and Lda2Vec], dataset [abstract and full-text], and pre- processing methods [WordNet Lemmatizer, Noun + Verb, and Noun + Verb + Adjective] over each of the 7 journals using jaccard distance to plot and visualize the correlation heatmaps. [7 journals × 2 datasets × 6C2 = 210 comparisons]

Responsibilities:

• Developed the Student’s Union App
• Guiding and Leading the App Development Team

Worked directly under Dr. S. Ramachandran, Senior Principal Scientist, Council of Scientific and Industrial Research - Institute of Genomics and Integrative Biology, Delhi, India.

• Project was on clustering and visualizing similar medical terms present in biomedical literature. Scrapped over 7074 abstracts on the topic of “Vitiligo” from PubMed. Handled entire end-to-end pipeline of data acquisition, data pre- processing, model training, and visualization of results.
• The data pre-processing was handled using regular expressions, stop-word removal, and lemmatisation. Bi-grams from the data were obtained using Gensim, and the common data pre-processing stages were implemented using NLTK.
• GloVe model was then trained on the dataset to obtain vector embeddings of the words and bi-grams. Visualization of the embeddings was done using Gephi. Thus, the vocabulary terms were visualised in the form of a weighted graph, where the weight of the edges between two terms was equal to the cosine similarity of their respective embeddings. Project on Github.