Data Scientist

Education

Received Bronze medal amongst 200 students by leading a team of 3 members for this 2-week long Datathon using advanced machine learning.
Achieved F1-score of 0.86533 by incorporating all the steps of the data science solution lifecycle to predict student dropout from courses

Utilized technologies such as Langchain, LlamaIndex, OpenAI API and the PapersWithCode API to develop an RAG application for researchers.
Stored indices in a vector database called Pinecone for an LLM Gemini Pro to effectively retrieve information about the latest AI in mental health research papers.

RAG Application

Developed an end-to-end ML solution to predict whether an existing insurance beneficiary will purchase vehicle insurance.
Utilized various Machine learning algorithms like logistic regression, decision trees, random forest, XGBoost classifier and LGBM classifier to select the best-performing model and later deploy the solution on Azure cloud.

Insurance predictor

Engineered image analyzing tool to provide additional data points about the image by using Instructional tuned LLM.
Utilized Google Gemini Pro API to conduct prompt engineering and achieve desired results.

Gemini-Vision

Performed exploratory data analysis about job descriptions from 2020-2023.
Completed thorough analysis using R to find out the top 10 job titles for an aspirant targeting entry-level full-time jobs.

Job Title

Reduced ~$100K in expenses by building interactive Power BI dashboards aiding internal and external stakeholders to make informed decisions.
Single-handedly developed the firm’s entire database by creating an API with a third-party service to schedule batch import at frequent intervals.

Saved ~$3.5 million in cost by automating IT tasks/processes by building an intelligent automation platform and saved 3+ hours of manual labour for every client with the help of opted services.
Implemented and maintained APIs and services using Java in collaboration with the UI team which utilized HTML, CSS and JavaScript for the platform to function smoothly.
Secured a Runner-up position in NeuroHack, an organization-wide hackathon, while leading a 5-member team to create an AI-driven portal for IT ticket categorization and anomaly detection.
Developed NLP solution and trained BERT model for generating ticket category-2 and incorporated HDBSCAN to generate ticket category-1 from the dataset.

Developed CNN – Convolutional Neural Network classification model with an accuracy of 96% to differentiate between pizzas.
Designed an instance segmentation model for biomedical images with an accuracy of 89% (optic disc and fovea detection) and documented a well-designed report and effectively communicated overall results and conclusions of the project.