Projects
MSR: Markovian Semantic Relatedness
Demo coming soon
Unsupervised query expansion and cross-language search. MSR builds a term
co-occurrence graph, runs a Markov-chain walk toward a root term, and scores every
term by z-scored hitting time. The result is a weighted concept that can rank
untranslated foreign-language text for an English query, without machine translation.
I designed MSR and led the project as a U.S. Government researcher (2019–2022). This is
a from-scratch reimplementation.
Python · NumPy/SciPy · spaCy · FastAPI · Celery · RabbitMQ · PostgreSQL · Next.js · Docker
CHIP: Confounding Hash Iteration for Passwords
Live demo
An iterated password hash that keeps hashing
until it has seen a set number of collisions among the low bits of its digests, so the
iteration count depends on the password and salt in a way you can't predict without
running the whole chain.
Python · Pyodide · Web Workers
What I've worked on
- AI agents. Request-routing agents, custom function tools, file retrieval, integrations, automations, and a text extraction pipeline for an AI executive assistant.
- Zero-knowledge proving. Invented the mathematical approach behind Sagittal MSM and helped build a multi-cloud proving platform that ran GPU workloads for production customers.
- Blockchain analytics. Python and Django services for token and NFT event analysis, cryptocurrency price ingestion, and wallet segmentation.
- Confidential AI. Prototyped encrypted client-to-enclave requests for model inference in a trusted execution environment.
- Semantic search research. Led a team building Markovian Semantic Relatedness, a cross-language retrieval system using unsupervised learning and graph analytics.
- Computer vision research. A real-time system that used a high-speed camera and CUDA to recover sound from the vibrations of ordinary objects.
Education
The George Washington University
M.S. in Computer Science
B.S. in Computer Science, minors in Mathematics and Music
Fun fact: I've performed for half a million people on trumpet 🎺
