[featured] mar 2026
deep learning dog breed classification
custom pytorch cnn and vision transformer architectures for 10-class dog breed image classification across a 9,000-image dataset.
- implemented transformer encoders, multi-head self-attention, and scaled dot-product attention
- built transfer learning pipelines with adam optimization, checkpoint serialization, tensor normalization, and 256-patch embeddings, training models over 10,000+ update iterations
pytorch cnns vision transformers python
march 2026
personal website
personal portfolio built with modern web tooling. responsive layout, scroll and interaction-driven animation, and structured sections for experience, projects, and contact.
- responsive design across breakpoints
- animation-heavy ui with framer motion
- structured content sections with clear hierarchy
next.js typescript tailwind css framer motion
github
feb 2026
icu mortality prediction model
clinical ml pipeline predicting icu mortality from sparse ehr time-series data across 12,000+ admissions and 40+ physiological variables.
- transformed sparse ehr time-series data through normalization, imputation, statistical aggregation, and feature engineering workflows
- executed 5-fold cross-validation and 1,000+ bootstrap resampling iterations across logistic regression, kernel ridge regression, and rbf kernel models, benchmarking auroc, sensitivity, and specificity under imbalance conditions
python scikit-learn numpy pandas
nov 2025
sudoku solver
c++ solver for 9x9 sudoku. represents the board as a 9x9 grid, validates rows, columns, and 3x3 boxes, tracks candidate values for open cells, and solves via repeated constraint checks and possibility elimination.
- 81-cell board with row, column, and box validity checks
- candidate tracking and constraint propagation
- solves nyt easy puzzles in under a second; runtime measured with chrono
c++ stl chrono
github
oct 2025
order book simulator
price-time-priority central limit order book matching engine in c++ using priority queue (heap-backed) bid/ask books for low-latency trade execution.
- processed 1m+ orders and executed 760k+ trades in under 10 seconds (74k+ orders/s) with o(log n) average matching complexity
- implemented limit and market order types with real-time bid-ask spread tracking, partial fill logic, and order cancellation, validating engine accuracy against 500k+ expected trade outputs with automated test scripts
c++ heaps data structures
june 2025
bst-based map container
binary search tree with sorting invariants and an ordered map adt for efficient key-value storage.
- sorting invariants, traversal logic, functors, templates, and recursion
- o(log n) insertion and lookup performance
c++ bst templates
may 2025
naive bayes text classifier
multivariate bernoulli naive bayes classifier to classify posts by topic using log-probability scores.
- determined labels using highest log-probability score
- abstract data types for efficient file parsing and word-frequency detection
- generates per-label log-likelihood maps used for automated topic classification
c++ probability adts
oct 2021
tic-tac-toe
interactive tic-tac-toe game in python using turtle graphics. keyboard-controlled for both players.
- board, xs, and os drawn with turtle; win and tie detection
- scoreboard, reset, and replay with basic error checking
python turtle
github