End-to-end, open & open. Until every end-to-end flow is open to public.

I do about almost everything that interests me. Im not kidding.

GitHub page

small projects from time to time; most big projects are in orgs. I like useful software, silly ideas that somehow work, and small projects that teach lessons.

personal

team

OSS

• local serving & inference - running and evaluating models locally.
• applied ml & dl - classification, recommendations, and applied models.
• federated learning - experimenting with capability isolation and access to shared models.
• dllms - exploring diffusion language models and comparing them with autoregressive models.
• model behavior & evaluation - reproducible experiments on bias, reliability, and failure modes.

• agentic workflows - tool use, multi-step execution, and coding agents.
• mcp, plugins & skills - connecting agents to tools and extending their behavior.
• tool selection & trust - studying how agents choose between legitimate, malicious, and decoy tools.
• privacy policy & information flow - tracking sensitive data across tools, memory, and agent actions.
• prompt injection - measuring attacks and defenses across languages and tool surfaces.

• cli & tui - terminal-first tools and interfaces.
• git & contribution tooling - version control, review workflows, and open-source contribution tools.
• agent infrastructure - building, extending, and evaluating coding agents.

• linux installation & recovery - boot flows, dual boot, rollback, and hardware support.
• test automation - vm-based installation scenarios and system validation.
• custom os - exploring practical desktop systems and lower-level software.
• native & cross-platform software - building tools that run across windows, macos, and linux.
• fuzzing - experimenting with harness generation and execution feedback, including dllm-guided approaches.

• web apps & services - applications, apis, bots, and community tools.
• games & simulations - playable systems and interactive experiments.
• terminal graphics - rendering games, video, and 3d scenes in a terminal.

• ai: python, pytorch, transformers, vllm, scikit-learn, krasis
• agents: mcp, opencode, claude code, agent skills, playwright, codex
• dev tools: rust, go, typescript, bun, opentui, git, github actions
• low level: linux, docker, qemu, clang, asan, ubsan, wasm
• web & apps: react, next.js, three.js, fastapi, postgresql, redis
• research: pytest, coverage tools, latex, jupyter

I care about open source because it turns private learning into public infrastructure. I mostly gravitate toward developer tools, AI infrastructure, and software people can actually run and inspect.

If you want to collaborate, talk shop, coffee chat, or build something questionable in a productive way:

jay7math@gmail.com

Not interested in:

• crypto miracles
• “star my repos and back”
• “definitely not a malware!!! pls PR this .exe!!!”
• “yo this massive security vulnerability of a program is the new trend try it out”

english: Literally the preferred choice in many scenarios. (OPIc AL)
korean: Native language; I do poetry and short stories occasionally.
chinese: Relearning (former YCT4).

GitHub · LinkedIn · Naver · Velog · Ko-fi