projects

Amalgamation of various short and long-term undertakings.

machine learning

▸
[Re] FlashAttention — Triton implementation of FlashAttention-1 with Multi-Query Attention (MQA).
code · paper · poster
▸
LLM Homogenization — With sufficient initial context, LLM-generated text corpora achieve stylistic and semantic diversity on par with human writers.
paper
▸
Astralis — RAG pipelines for geospatial real estate AI. Built intelligent document ingestion using Python, LLMs, and ML clustering.
code · site

kernels & numerics

▸
[Re] Spectral Methods — Ultraspherical spectral solver from Olver and Townsend (2013).
code · paper
▸
GPU MODE — Competition Triton kernels for GPU MODE.
▸
Butterfly Matrix Kernel — Structured butterfly matrix multiplication transforms in Triton.
code

hackathons

▸
In-Context Scheming — Multi-agent LLM corruption propagation experiments.
2nd place, Anthropic Alignment Hackathon 2025
▸
SearchDestroy — Multi-drone area sweep with online re-planning via DARP + heuristics.
2nd place, AGI House Robotics Hackathon 2024
code

bio / chem / neuro

▸
c-Myc mRNA Binders — Covalent binders for cancer-linked RNA at Disney Lab (UF Scripps). Presented at Max Planck Florida.
abstract