← Home

Selected Work

Things I’ve built.

Search infrastructure in Rust, graph learning for financial crime, and engines written close to the metal.

01

Contractor Marketplace

Aug 2025 – Present
  • Rust 1.97
  • PostgreSQL / PostGIS
  • React 19
  • GCP
  • Built a Rust search backend on GCP for ~50,000 licensed contractors, combining full-text search, location filtering, and natural-language trade matching.
  • Built data pipelines to compile California contractor records from CSLB and enrich them with Google reviews using Apify, linking licensing data with customer feedback.
  • Integrated production Firebase Authentication for Google sign-ins, and Argon2id password hashing for email sign-ins.
Visit site →
02

Subgraph Matching — Directed Research

Mar 2026 – Jun 2026
  • C++20
  • Knowledge Graphs
  • Branch & Bound
  • Built algorithms for inexact attributed matching over knowledge graphs, scoring node and edge label correspondences by embedding similarity. Applied to DARPA datasets to surface laundering typologies in large, noisy financial networks.
  • Adapts Tu et al.’s branch-and-bound subgraph matching solver by implementing it in C++ and adding structural equivalence pruning to avoid redundant search branches.
GitHub ↗
Structural equivalence in a query graph Clustered structural equivalence classes
03

GNN for Ethereum AML Detection

Jan 2026 – Mar 2026
  • Rust 1.90
  • Python 3.10
  • Graph Neural Nets
  • Transfer Learning
  • Adapted IBM’s Multi-GNN architecture from edge-level Bitcoin AML classification to node-level Ethereum sanctions detection, learning node embeddings via transfer learning against a 14,140:1 class imbalance.
  • Engineered a Rust pipeline with crash recovery to collect (via Alchemy APIs), label, and aggregate 90M+ transactions across 33M+ addresses into weekly directed graphs in Parquet.
Training ↗ Viz ↗ Preprocessing ↗
GNN AML visualization GNN AML visualization GNN AML visualization
04

Typing Tutor

Mar 2024 – Jun 2024
  • x86 Assembly
  • Irvine32
  • Built a falling-word typing game in assembly, tracking completed words and ending gameplay when a word reaches the bottom of the screen. Low-level design was implemented around the CISC x86 architecture.
  • Managed game state and word pointers in memory, implemented stack-based procedure calls with the designated stack pointer (ESP), and integrated Irvine32 routines for console input and output.
Try demo → GitHub ↗
Typing Tutor demo
05

B+Tree SQL Engine

Mar 2024 – Jun 2024
  • C++11
  • B+Tree
  • SQL Parser
  • Built a relational database engine from scratch, including a SQL parser (state machine implementation) and a query execution pipeline supporting SELECT, INSERT, and conditional filtering (Shunting-Yard / Boolean RPN).
  • Achieves logarithmic retrieval complexity through a unique B-plus-tree and dictionary database, using standard 1D arrays and binary (.bin) files for low-latency data retrieval.
Try demo → GitHub ↗
B+Tree SQL demo