// featured
FEATURED · ML
Neural Network from Scratch
Complete deep learning library in pure NumPy — no PyTorch, no TensorFlow. Forward propagation, backprop via chain rule, SGD / Adam / RMSProp, batch normalization, dropout. Built to understand exactly what happens inside the black box, starting from ∂L/∂w.
// quant & stochastic
01
SDE Path Simulator
Monte Carlo simulation of GBM, Ornstein-Uhlenbeck, and Heston models. European and Asian option pricing with variance reduction (antithetic, control variates).
↗
Δ · Γ · Θ · ρ
option greeks
02
Black-Scholes Calculator
Full BS option pricer with Greeks (Δ, Γ, Θ, ρ, ν), implied volatility surface, and interactive payoff diagrams. Derived directly from the heat equation via Feynman-Kac.
↗
03
Gradient Descent Visualizer
Interactive 3D loss surface explorer. Animate SGD, Adam, RMSProp trajectories on real loss landscapes with adjustable hyperparameters. WebGL + Three.js.
↗
Attention(Q,K,V)
= softmax(QKᵀ/√d)V
04
Transformer from Scratch
Multi-head self-attention, positional encoding, FFN layers, layer normalization — implemented from Vaswani et al. 2017's math. No high-level APIs.
↗
claim → warrant → impact
argument structure parser
05
Debate Evidence Analyzer
NLP tool that parses policy debate evidence cards, extracts argument structure (claim / warrant / impact), and scores quality using transformer sentence embeddings.
↗
06
K-Means from Scratch
Unsupervised clustering with animated convergence. K-means++, elbow method, silhouette analysis. Includes a convergence proof via the alternating minimization framework.
↗
this site
07
Personal Site
This portfolio — pure HTML/CSS/JS, no frameworks. Brownian motion background canvas, typewriter effect, animated GBM chart, skill bars, neural network animation.
↗
// more projects in progress — check back