//Projects

A TypeScript framework for building interactive terminal user interfaces with 1,500+ downloads on npm. Converts declarative TypeScript into fully interactive TUI applications runnable via a single npx command. Features 30+ pre-built UI components, a 12-column responsive grid system, spatial arrow-key navigation, 10 color themes, an ASCII art engine, file-based routing, reactive state management, and form handling. Minimal dependencies — just esbuild and Node.js built-ins. Includes 2,185+ tests.

A platform connecting computational neuroscience with social simulation for pre-launch advertising analysis. Predicts how ads will perform neurologically and socially before launch by combining Meta TRIBE v2 brain encoding (trained on 450 hours of fMRI data) with MiroFish/OASIS multi-agent social simulation (200 personality-diverse agents). Features Neo4j GraphRAG integrating 22 neuroscience papers, LangGraph diagnostic analysis, and Three.js brain visualization. Built with Next.js 15, React 19, FastAPI, and OpenRouter.

Served as the Lead Developer for a youth-led NGO focused on providing education to underprivileged children and fostering a passion for learning among youth.

Designed a time series model using LSTMs to predict global temperatures, aiding planning and combating climate change.

A web application that generates realistic handwritten text using deep learning. Features two neural network approaches: a 3-layer LSTM (14MB, hidden size 400) with 20 bivariate Gaussians and K=10 soft Gaussian attention windows for style transfer, and a 6-layer cross-attention GPT decoder (102MB, d_model=384, 6 attention heads) with a polar tokenizer using 128 angle × 64 radius bins. Supports up to 10 custom handwriting styles. Built with PyTorch, FastAPI, and JavaScript.

Created an AI-driven jigsaw puzzle solver that automatically assembles pieces using computer vision techniques.