Portfolio

This summer, I worked at Mobius Labs, a very early-stage startup based in Menlo Park, California. Mobius is leveraging tools that allow corporations to integrate generative AI into their business processes. I worked on fine-tuning generative AI models such as text-to-image models and open source LLMs.

I was also responsible for staying up to date on the latest technologies in the AI space that could be valuable to the company’s products. I experimented with emerging AI technologies such as LLM plugins and tools.

Stuff I’m worked on:

• Building data analytics tools with LLMs
• Testing viability of LLM tools to become potential products
• Deploying models on AWS to run open source LLMs (text-generation-webui, Sagemaker, custom AMIs on EC2)
• LangChain, Q&A chatbots, and agents
• Building chat interfaces with Docker, React, and FastAPI
• Constructing demos for potential enterprise customers using Streamlit
• Getting product demos ready for enterprise contracts and VC funding

I worked within the Data Science Division - MLOps Team at Mailchimp/Intuit. I experimented with methods to optimize the deployment of data science models to production.

I audited three of the datascience pipelines and proposed a myriad of problems and solutions relating to documentation, testing, versioning, notebook development, and production readiness. I was responsible for automating Python and SQL documentation generation and version control. I integrated the solutions into our new repository templates.

Some of the cool technologies I’m worked with and learned:

• Kubernetes and Docker
• Jenkins and Airflow
• Python Library Template Engines
• Google Cloud Platform

I am a part of the QUEST (Quality Enhancement Systems & Teams) Honors Program in Cohort 40. QUEST is an interdisciplinary program for business, engineering, and science students to gain hands-on experience within the industry by partnering with large corporations. We learn how to communicate with clients, create solutions to real problems, and work within intellectually diverse teams.

For the first course, my team and I designed Verbalinkd, a platform to connect immigrant restaurant owners connect with service professionals that they can communicate with. A large portion of immigrant entrepreneurs struggle with language barriers on a daily basis, and we aimed to alleviate them. Read more about that here!

My group consulted for the University of Maryland’s Engineering Career Services, where we designed potential solutions to help alleviate issues relating to registration and check-in for employers at our school’s career fairs. Read a detailed report about our story here!

Our curriculum consists of the following:

• BMGT 190H: Introduction to Quality and Design
• BMGT 438A: Applied Quantitative Analysis
• BMGT 390H: Designing Innovative Systems
• BMGT 490H: A Capstone project where students consult QUEST’s coporate partners on an identified organizational challenge

We teach public forum debate to middle school students around the country. I have coached students on how to plan debate strategies, organize evidence, think critically, and speak with authority. I handled Trailblazer’s sign-up, administrative organization, advertising, and financial system. Our team has successfully hosted 6 debate camp sessions so far!

Our Curriculum:

• Public Forum Debate Fundamentals
• Basics of Argument
• Public Speaking, Rhetoric, and Clarity
• Research, Writing, and Analysis
• Notetaking, Responses, and Strategy

I conducted a research project at what would be my future alma mater in my junior year of high school. Having exprimented with Twitter’s Developer API before my internship, I was inspired to do machine learning research on public sentiment on the social media platform. I worked under Dr. Daniel Lathrop and Dr. Heidi Komkov of the Nonlinear Dynamics Laboratory.

• Designed an Artificial Neural Network with Pandas, Scikit-Learn, and TensorFlow to predict the exposure of tweets using sentiment analysis.
• Created a custom data pipeline using Python and the Twitter Application Programming Interface.
• Collaborated with mentors to determine and optimize hyperparameters for making predictive estimations of exposure.
• Presented research to laboratory and peers.

I and four other collaborators studied the effects of gene expression during embryogenis on cancer metastasis. We queried the E14.5 mouse embryo database using CBioPortal, one of the leading online hosts of cancer research data.

• Queried and identified a dozen critical cancer-related genes in the central nervous system of the E14.5 mouse embryo
• Performed statistical analysis and analyzed Kaplan-Meier Curves for each gene using cBioPortal genomic datasets

Our abstract was accepted by the AACR, and we were invited to present at their annual conference in 2020. Unfortunately, due to the COVID pandemic, we were not able to present.