Tang Ziya tcztzy

I build research software at the intersection of computational biology, LLM systems, and developer tooling.

I am Tang Ziya (唐梓涯). Tang is my family name, Ziya is my given name, and tcztzy is the initialism of 天才之唐梓涯 (Genius Tang Ziya).

I care about software that makes scientific work more legible, reproducible, and composable.

Working on agent skills, MCP servers, model-serving infrastructure, and scientific reasoning workflows.
Interested in research engineer, ML systems, and computational biology opportunities in Europe, the UK, and the US.
Usually shipping with Python, Rust, Jupyter, Nix, uv, and GPU/HPC-heavy workflows.

Selected Work

skills Installable agent skills for Codex and Claude, with marketplace generation and local sync tooling.	swarmx A lightweight multi-agent orchestration framework with MCP support and an OpenAI-compatible server.
yallm A high-performance API hub that maps between clients and multiple model providers with streaming-first compatibility.	xcc A modern Python C compiler focused on deterministic behavior, transparent diagnostics, and a clean architecture.
RL4GenomeBench Benchmarking and reinforcement learning for scientific reasoning in genomics and CRISPR-related tasks.	cotton2k A Rust rewrite of an established cotton simulation model with Python bindings and batch workflows.
JupyMCP An MCP server for rich Jupyter notebook interaction, including cell editing, execution, and notebook resources.	ZotMCP A bridge between Zotero libraries and AI assistants through MCP for retrieval, annotation, and literature workflows.

Hardening yallm and swarmx around provider compatibility, streaming, and MCP-native workflows.
Building research tooling for notebooks, libraries, and local agent environments.
Exploring scientific reasoning, evaluation, and RL training in genomics-focused settings.
Keeping one foot in systems work through compiler and infrastructure projects like xcc.