███████╗ ██████╗ ██╗  ██╗ █████╗ ██╗██████╗
 ╚══███╔╝██╔═══██╗██║  ██║██╔══██╗██║██╔══██╗
   ███╔╝ ██║   ██║███████║███████║██║██████╔╝
  ███╔╝  ██║   ██║██╔══██║██╔══██║██║██╔══██╗
 ███████╗╚██████╔╝██║  ██║██║  ██║██║██████╔╝
 ╚══════╝ ╚═════╝ ╚═╝  ╚═╝╚═╝  ╚═╝╚═╝╚═════╝

Still a student, but I'm one of the core contributors on semantica, a knowledge-graph/RAG framework , mostly poking at the graph store and export side (Neo4j, pgvector, Apache AGE) plus some work on the React frontend.

Outside of that I mess around with computer vision and RL a lot, and lately I've been building fintrack-graphrag, a pipeline that pulls SEC 10-K filings apart and makes them queryable.

• fintrack-graphrag – rips apart SEC EDGAR 10-K filings, embeds them, lets you actually ask questions about them. TS/Rust/Java/Python, was a good excuse to write Rust for once
• AutoServe – serving YOLO models fast with FastAPI + Triton, tried to squeeze out as much latency as I could with CUDA shared memory
• ClaimTrace – tries to catch deepfakes / mismatched image-caption pairs using VQA + NLI + ELA, started as a "can this even work" experiment
• FraudGuard – autoencoder watching a stream of transactions for anomalies

PyTorch, FastAPI, Neo4j, Postgres (pgvector / Apache AGE), React, Java and whatever Jupyter notebook I left open last. Picked up Rust while building fintrack-graphrag and liked it more than I expected to.