Contributed to a teen suicide prevention app that provides teens with tools to adhere to their safety plans. Implemented a mood tracking system, a culturally sensitive resource system, integrated a Calendly appointment scheduling & tracking system, translated the app in 4 languages, and improved app loading times by 200%. Leveraged AWS Lambda to apply styled HTML/CSS templates to the app's emails in 4 languages. Leveraged AWS EventBridge & Lambda to create 4 recurring hydration jobs.

Fine-tuned & compared 9 GPT & Bert models in terms of accuracy when answering biomedical questions from the PubMedQA dataset. Models included GPT-2, BioGPT, BioBERT, BERT, Clinical-BERT, DistilBERT, MedBERT, PubMedBERT, & BlueBERT. Performed experimentation with varying training splits to highlight the models' few-shot learning capabiltities. Results indicated that PubMedBERT was the best performing model boasting a mean accuracy of 84%.

Created a pal app for pets in which pet owners can create profiles for their pets to connect with others. Pet owners can create profiles for their pets, swipe through pets, match with pets, and view their matches. I facilitated the app's API routes, page routing, session management, image rendering, database queries, and hosting / deployment. Fulfilled the positions of PM and Tech Lead by managing a team of 6 developers and leading development.

Conducted linear regression analysis on Virginia's Backbay Refuge water quality data to determine the relationship between water quality predictors and dissolved oxygen levels. Conducted exploratory data analysis to identify key predictors and potential outliers. Performed backward elimination during model selection to identify the best predictors. The model was trained on half of the data and tested on the remaining half of the data. The final model produced an R-squared of 0.6 and MAE 2 mg/L.

Developed a personal portfolio to showcase myself along with my skills, work experience, and projects. Leveraged TypeScript, React, Next.js, Shadcn, DaisyUI, and TailwindCSS to create the portfolio. The portfolio is hosted using Vercel. The portfolio is responsive, snappy, and mobile-friendly while following a minimalistic design.

Implemented a social platform that facilitated events and club administration for universities. Enabled students to share event/club info such as their locations, contact info, and memos. Ensured that only registered students with emails were allowed to access this social platform.

Developed a contact manager using the LAMP stack. Contastic allows users to add, delete, and update contacts. Implemented a search feature to search for contacts by any attribute. The contact manager was hosted on a Digital Ocean droplet. Ensured security by protecting against SQL injection attacks.

Built and trained an LSTM Model to conduct sentiment analysis on the IMDB movie review sentiment dataset. Leveraged pytorch to build an LSTM model with BCE Loss and with 1 embedding, LSTM, and linear layer. Trained the LSTM model in batches for 10 epochs resulting in an average of 85% validation accuracy and 0.2 loss.