A selection of projects in machine learning, data analysis, numerical methods, and self-hosted infrastructure.
DataLens
Python · Streamlit · Pandas · NumPy · scikit-learn · XGBoost · Docker · pytest
Built a tool that profiles unfamiliar tabular data, identifies credible targets and analysis paths, and avoids forcing predictive models onto weak use cases.
- Implemented leakage-safe preprocessing and cross-validation across linear, ensemble, and XGBoost candidates.
- Added baseline comparisons, drift checks, and external validation to test whether results generalize.
- Routes weak predictive cases to evidence-backed descriptive analysis instead of recommending an unreliable deployment.
View on GitHub
CareerAgent
Python · FastAPI · PostgreSQL · Next.js · Docker · Playwright · LaTeX
Built a human-in-the-loop workflow that discovers, verifies, ranks, and tracks job opportunities and generates review-ready application packets.
- Created explainable scoring for fit, freshness, availability, and close risk.
- Reserved LLMs for explicit drafting requests rather than using them for every step.
- Added review-gated Playwright autofill that never submits applications automatically.
View on GitHub
CIFAR-10 Image Classification
PyTorch · torchvision · CUDA · NumPy
Built a GPU-accelerated training and evaluation pipeline for UC Berkeley Data 188 and implemented a residual convolutional neural network.
- Improved test accuracy by 14.8 percentage points over an MLP baseline.
- Reduced the parameter count by 92% while improving classification performance.
Spam Email Classification
Python · scikit-learn · Pandas · Matplotlib
Built a logistic regression-based classifier to separate spam and non-spam emails using custom-engineered features.
- Designed features such as word count, character count, HTML tag presence, capitalization ratio, and spam-word frequency.
- Used exploratory analysis to identify which engineered features were most useful for classification.
- Trained and evaluated a logistic regression model using ROC curves and iterative feature improvements.
- Improved model performance through feature refinement and stronger signal selection.
Homelab Infrastructure and Self-Hosted Services
Proxmox · Linux · Docker · WireGuard · AdGuard · Nextcloud · Immich
Designed and maintained a self-hosted server environment for secure remote access, storage, photo hosting, and internal network services.
- Configured virtual machines and containers in Proxmox to run multiple services in a stable Linux environment.
- Set up WireGuard VPN and internal-only access patterns to improve security and reduce public exposure.
- Managed deployments for services including Nextcloud, Immich, and AdGuard Home.
- Built practical experience with storage, networking, reliability, and systems administration.
Modified Zeroin Algorithm Implementation
MATLAB · Numerical Methods · Root Finding
Implemented a hybrid numerical root-finding algorithm combining the Bisection Method and Inverse Quadratic Interpolation.
- Built logic to switch methods when interpolation was unreliable or converged too slowly.
- Tested the algorithm on benchmark functions under different tolerance levels.
- Improved understanding of convergence, interval methods, and numerical error behavior.
- Strengthened MATLAB implementation and debugging skills through hands-on mathematical programming.