Raised 200k pre-seed to engineer a multi-tenant SaaS platform empowering users to instantly launch, customize, and manage e-commerce storefronts on unique subdomains, with real-time AI-driven site/page/product generation.

Integrated Stripe for secure, multi-variant payments and automated tax calculation, orchestrated serverless deployment with 99.9% uptime on Vercel and implemented Redis caching for faster storefront load times.

Engineered high-performance implementations of Karatsuba and 3-way Toom–Cook multiplication algorithms for arbitrary-precision integers, leveraging cache-aware data structures and optimized vectorized digit arithmetic.

Designed novel parallelization of 3-way Toom–Cook using OpenMP and ParlayLib primitives, exploiting divide and conquer independence for fine-grained concurrency, achieved 535.25× speedup, with asymptotic scaling analysis.

I developed a price comparison web application and Chrome extension for finding related products at a lower price with optimized result caching using Redis for fast, repeatable user queries with affiliate link APIs to monetize referrals. I implemented AI-driven product matching using natural language embeddings and fuzzy search and Firecrawl for dynamic web crawling and HTML parsing to extract product metadata and pricing across multiple vendors.

I utilized an interpretable model to quantify and confirm the cerebellum's importance in predicting risk tolerance and evaluated the performance of several machine learning techniques using 1000 runs of 10-fold cross-validation. I published work using machine learning to assess the relationship between risk tolerance and brain gray matter volume as first author, and presented at the Organization for Human Brain Mapping (OHBM) Conference 2023.

I created a full-stack web app to track workouts, visualize progress, and integrated AI insights for better training. I deployed the app on Render, serving 100+ users and 10,000+ data points through clean UI and optimized backend.

Log page:

Chat page:

This was quite challenging since I was not altogether too familiar with the codebase or with the packages and tools we used in this project, so it took a while for me to get the project off the ground, but it was quite fun in the end since the simulator was extremely visual, so it was easy to see what was happening and how all of the code was affecting the environment. This was also quite rewarding because I was able to see that the robot actually learned the policy from the training, and although it was a simple task, it solidified our results.