research archive

Current investigations and completed research across physics, machine learning, and genomics.

This section tracks active experiments and finished work, with an emphasis on computational methods, interactive tooling, and scientific interpretation.

active now

Projects currently being extended or validated.

completed

Finished research work kept online for reference.

focus areas

Distinct themes spanning theory, data, and tooling.

research archive

current focus

Anomaly Detection in SDSS Stellar Spectra

Seven anomaly detectors — from Isolation Forest to a conditional normalizing flow — running on public SDSS DR18 stellar spectra. Includes a physics-motivated PBH candidate screening pipeline, semi-synthetic evaluation harness, and a Streamlit dashboard. Scales to 50K+ spectra.

AstrophysicsMachine LearningAnomaly DetectionPythonDeep LearningSDSS

why it stands out

The active work now mixes scientific validation with implementation detail, which gives the portfolio a stronger “research in motion” signal than a static archive alone.

Open write-up → External resource →

Anomaly Detection Astrophysics Attention Bioinformatics CUDA Computational Physics Data Science Deep Learning GPT-2 GPU LLM Machine Learning Mean-Field Theory Monte Carlo PCA Physics Pruning PyTorch Python Quantization Research Rust SDSS Time Series TimescaleDB Trading WebAssembly XGBoost

research archive

XGBoost Signal Confidence Filter for Algorithmic Trading

An XGBoost classifier trained on 3,033 historical signal outcomes that doubles win rate in a crypto trading system — from 24% to ~50% — by scoring each signal's confidence before execution. The filter worked; four months of dry-run evidence then killed most of the strategies it was filtering. Post-mortem included.

Machine Learning XGBoost Trading Time Series Python TimescaleDB

completed repo linked →

research archive

Physics-Inspired LLM Optimization: 5 Directions, 9 Experiments

Testing whether structures from physics — thermodynamics, renormalization group, lattice theory, and mean-field dynamics — can guide LLM compression. Five research directions validated on GPT-2 124M: entropy-based quantization (2.6× better than uniform), RG-guided pruning, lattice attention with a phase transition, GPTQ interaction analysis, and metastable cluster verification.

Physics LLM Quantization Pruning Attention Mean-Field Theory GPT-2 PyTorch

completed repo linked →

current investigation

Anomaly Detection in SDSS Stellar Spectra

Astrophysics Machine Learning Anomaly Detection Python Deep Learning SDSS

active repo linked →

research archive

Physics-Inspired k-Means Quantization for LLM Weights

26 experiments compressing GPT-2 with physics-motivated k-means codebook quantization. Best result: PPL=84.2 at 0.836 bits per weight — 38× compression vs FP32. Null results triangulate the GPTQ design from first principles.

Machine Learning LLM Quantization Physics Python GPT-2 Research

completed repo linked →

current investigation

Ising Model Simulation in Rust + WebAssembly

A publication-grade Monte Carlo simulation engine for classical spin models, achieving 0.01% accuracy on the 3D critical temperature. Supports three universality classes, GPU acceleration via CUDA, and finite-size scaling analysis — targeting Physical Review E.

Physics Rust WebAssembly Monte Carlo Computational Physics CUDA GPU

active repo linked →

research archive

Genetic Substructure Analysis via PCA

Analysis of genetic substructure across global populations using genome-wide SNPs and principal component analysis on the 1000 Genomes Project dataset.

Bioinformatics Python PCA Data Science

completed read more →