Optimizing inference latency and deployment of source code detection models for distributed large-scale environments

I engineered a full-stack solution from data acquisition to model deployment to personalize the shopping experience through a tinder-like experience. I built an automated Python-based web scraping pipeline using Playwright to populate a PostgreSQL database. The core of the project is a sophisticated recommendation model that uses multi-modal (image and text) embeddings from a CLIP-based transformer. To deliver highly relevant and engaging suggestions, I implemented advanced features including negative feedback modeling, session-based personalization, and diversity-aware re-ranking with Maximal Marginal Relevance.

As an AI/ML Intern at NASA, I am developing a custom LLM solution, integrating a Retrieval-Augmented Generation (RAG) system, to reinvent the content workflow for the MyNasaData platform, which brings NASA's Earths and Sciences resources into classrooms nationwide. This RAG-powered tool leverages prompt engineering to access a knowledge base of existing materials, enabling it to automatically generate new, grade-specific lesson plans. Additionally, it classifies and analyzes current lesson plans, providing educators with actionable recommendations to align content with federal educational standards, significantly streamlining the curation process.

I utilized prompt engineering to develop several specialized models with distinct capabilities. One model was trained to architect and solve sophisticated and novel puzzles in Python. Another was trained to interpret and author complex SQL queries, using a curriculum of advanced queries on production-style databases to ensure high accuracy and practical application.

During my internship, I contributed to an innovative AI tutor application designed to make learning more engaging. My main focus was on the front-end, where I used React to build key components of the fun, gamified interface for students. I also worked with Firebase to implement backend features for real-time progress tracking and user data management. A significant part of my role involved collaborating with the team to seamlessly connect the user interface with our sophisticated AI engine, which was powered by LangChain, OpenAI, and deployed on Azure.