Princeton Startup Immersion Program

• Built a semantic search tool using NLP for Indian languages with CTR of 23% and expected to serve ~400,000 queries/day

• Increased recall of search by ~10 times using a multilingual Sentence Transformer model and fine-tuning it on a custom dataset, which I created by analysing customer search queries and their interactions with items using SQL

• Developed Hindi audio summarization, topic modeling, and keyword extraction systems using LLMs, Term Frequency method, and LDA that automate categorization of customer calls and reduce call analysis time by ~75%

• Built a code commenting tool using GPT-3.5 API for the software teams to improve cross-team code reusability

Tigers in Product is the product/ product management subteam of the Princeton Entrepreneurship Club, with the goal to create product-minded thinkers and leaders on Princeton’s campus.

• Collaborated with potential speakers to design and organize product-themed workshops, expanding networking opportunities and knowledge-sharing within the community.

• Managed the internship cycle, conducting regular meetings to ensure a positive experience for interns and facilitating communication between interns and companies when needed.

• Organized a successful hands-on workshop on Product Management with Mr. Austin Mejia and an Intro to VCs session with Contrary, attracting over 300 participants.

• Recruited and effectively led a team of 20+ students, fostering a collaborative and productive environment.

• Oversaw and managed operations, marketing, and outreach teams to drive the club's initiatives and maximise impact.

• Conducted in-depth analysis of data pertaining to women's academic drop-out rates, aiming to identify underlying reasons and patterns.

• Employed Python to develop predictive models using a range of techniques, including linear regressions, support vector machines (SVMs), and neural networks.

• Performed image processing and built a similarity CNN for recyclables with 98% accuracy using OpenCV and Keras

• Built a business-critical recycling classifier that automates recyclable classification by using k-nearest neighbors and calculating image embeddings using specific layers of the trained similarity CNN, with 96% accuracy

• Refined dataset robustness by identifying classes with limited images through unlabelled image classification