Nice to meet you! 👋

I'm Tej

I'm a

Full-Stack Developer.

About

I'm a Computer Science and Linguistics student at the University of Maryland, passionate about developing innovative software solutions. I focus on leveraging AI, Machine Learning, and Data Science to solve complex problems across various industries, and I'm eager to explore new ways these technologies can be applied, uncovering creative solutions and pushing technological boundaries. Currently, I'm honing my skills as a Software Engineering Intern at Anoki AI, where I'm working on their multimodal ContextIQ platform for contextual video intelligence.

When I'm not coding, I enjoy playing soccer, golfing, working out, and spending time with my dog, Beau.

San Francisco, CA

Experience

Machine Learning Engineer Intern

Mountain View, CA

Jun 2025 - Present

•Improving the infrastructure, reliability, and scalability of Anoki's multimodal video analysis platform, ContextIQ, by migrating it from a single GPU server to Kubernetes.
•Building a Go library that uses the Kubernetes API to launch on-demand GPU jobs, triggered by media queries, to fetch video embeddings from S3, run the embedding service, and shut down automatically to optimize compute efficiency.

Computational Linguistics Research Assistant

College Park, MD

Jan 2025 - Jun 2025

•Collaborated with Dr. Naomi Feldman and graduate fellow Imani Finkley to design and execute experiments measuring textual novelty in LLM-generated literature, testing how prompt specificity affects model creativity.
•Applied the Creativity Index and n-gram analysis to measure lexical and thematic variation across model outputs.
•Fine-tuned and evaluated OPT-175b and BERT under varied prompt conditions to analyze LLM output behavior.

Software Engineer

College Park, MD

Jan 2025 - Present

•Working on a team to build a photo-sharing and scheduling platform for Camp Starfish, a nonprofit camp supporting youth with social-emotional challenges through a 1:1 staff-to-camper model.
•Built Firestore helper functions for employees, albums, and campers to safely handle multi-step database operations like creating, updating, and querying by ID, UID, or email, ensuring data consistency when multiple records must be changed together.
•Developed a responsive Login page with secure Google Authentication using Firebase, styled with Tailwind CSS, including dynamic error messaging for failed sign-ins and a fallback link for unregistered users.
•The system currently supports over 1,500 photos and 300 camper records each summer, used by 60+ staff to securely manage media and streamline camper scheduling.

Software Engineer Intern

Aug 2024 - May 2025

•Developed Kalai, a Natural Language Query Dashboard using Google Gemini and LangChain SQL agents to process plain English queries and return SQL responses, enabling users to retrieve data from Karya's datasets.
•Optimized backend SQL execution by introducing caching in PostgreSQL to reduce latency and improve performance.
•Implemented a role-based access control (RBAC) system to enforce dataset permissions, ensuring secure and restricted access to sensitive data.

Software Engineer Intern

Decision Factories

May 2024 - Aug 2024

•Built LiveReqs, an LLM-powered application that transforms high-level use case descriptions into detailed specifications, interactive flowcharts, and executable flows. The tool enables real-time validation and structured business feedback, and was presented to potential customers for feedback and refinement.
•Designed and developed the company website with a responsive user interface and clear navigation to communicate the company's mission, showcase projects, and feature product demonstrations.

Projects

✈️

MACHINE LEARNING

JourneyFlow: AI Travel Planner

Privacy-preserving iOS application that creates personalized travel itineraries using Apple's on-device foundation models and GPT-5, with deterministic, modular LLM reasoning powered by DSPy pipelines.

Swift

Apple Foundation Models

GPT-5

DSPy

💼

MACHINE LEARNING

QuickDesk: Freelancer Workflow Platform

Android application that automates freelancer workflows using Claude-powered agents to parse emails, extract project details, and generate contracts and invoices, supported by a React Native–FastAPI–MongoDB stack coordinating asynchronous Intake, Contract, and Billing agents.

React.js

FastAPI

Claude

MongoDB

☁️

MACHINE LEARNING

EcoNavix

AI-driven platform that optimizes supply chain routes using real-time environmental and traffic data, reducing carbon emissions by up to 35% while improving logistical efficiency.

Python

Flask

React.js

👥

SOFTWARE DEVELOPMENT

LiveReqs

LLM-powered application that enables Product Managers to convert high-level product requirements into detailed, testable use case documentation with visual flowcharts, executable code, and automated scenario validation.

React.js

Node.js

REST APIs

TLDraw

Claude

⭐

Full-Stack Development

Camp Starfish

Photo-sharing and scheduling platform for Camp Starfish, a nonprofit serving youth with social-emotional challenges. The system supports over 1,500 photos and 300 camper records each summer, used by 60+ staff to securely manage media and streamline camper scheduling.

Firebase

AWS

React.js

🚀

WEB DEVELOPMENT

CosmoSearch

Web application that lets users explore NASA's image library. Just enter a space-related term and date range (1920-2025) to browse stunning space images with titles, descriptions, and dates. Designed for quick, engaging cosmic discovery across any device.

Machine Learning

React.js

🧠

AI RESEARCH

Is GPT-3 Smarter Than A Sixth-Grader?

Research study that evaluated the GPT-3 Davinci model's ability to answer middle school science questions from the TQA dataset using zero-shot, few-shot, and fine-tuning approaches to measure performance across different learning setups.

Python

GPT-3

NLP

pandas

⚾

MACHINE LEARNING

WAR Prediction

Machine learning model built with Python and scikit-learn to predict 2019 MLB player Wins Above Replacement (WAR), a key metric for evaluating player value and informing team decision-making, achieving a 0.87 correlation with actual performance.

Python

scikit-learn

Linear Regression

🏟️

DATA SCIENCE

How To Get A Good Whiff

Logistic regression model built in R to predict optimal MLB pitch locations by analyzing swing-and-miss probabilities, achieving 85% accuracy using Statcast data and visualized insights through ggplot2.

ggplot2

dplyr

🐦‍⬛

MACHINE LEARNING

Anti Refugee Sentiment Analysis

Sentiment analysis model developed in Python using NLP techniques to classify tweets as pro- or anti-refugee, achieving 92.65% accuracy through feature extraction methods like Bag of Words, Tf-idf, and Word2Vec with logistic regression.

NLP

Python

scikit-learn

TF-IDF

Word2Vec