Senior software engineer · Distributed systems, performance and cryptography

I build and optimize systems where concurrency, correctness and privacy matter. My background includes open-source maintenance, peer-to-peer protocols, asynchronous Python services, data ingestion and encrypted synchronization.

Python has been my primary professional language, alongside project work in Erlang, Rust and Go. My research interests include GPU computing and hardware-aware optimization.

I was a core maintainer of the LBRY SDK, working on peer-to-peer networking, DHT lookup, encryption and metadata signatures. My professional experience also includes HyperBEAM contributions at Odysee and core maintainer of LEAP’s Soledad encrypted synchronization database.

gpu-vole explores CUDA implementations of FAEST and PERK post-quantum signatures. The repository includes KAT based correctness-test and benchmark tooling. It is research code, with the published article linked in there.; the README describes its reproduction requirements and security limitations.

cachelog-rs is a concurrent key-value map with write logging and caching. It includes model/proof and concurrency-testing artifacts, with an explicit boundary between the verified semantic core and the live concurrent implementation. It was made to answer "How well can LLMs work with formal proofs over data structures?" -- turns out, they did quite well.

VeryLogo / Space-Time Compiler experiments with Verilog-to-IR compilation and SIMD/PTX backends, which can be used by LLMs to auto research hot code paths. You describe the hot code path in Verilog and the tools driving lowering to a fast compiled version. better-compact explores VCC based context-compaction, session observability and search tooling for coding-agent workflows.