Skip to content
View Zymonody7's full-sized avatar
😍
nothing
😍
nothing

Highlights

  • Pro

Block or report Zymonody7

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Zymonody7/README.md
typing

MSc @ University of Chinese Academy of Sciences · Artificial Intelligence · Beijing

GitHub followers Visitors


🤖 What I do

  • AI Agents — my core focus: agent design, tool use, orchestration, and turning LLMs into reliable products
  • AI Infra — model quantization & inference optimization: HiFloat8 W8A8 PTQ/QAT, GPU kernels (MUSA GEMM, MoE routing on MetaX)
  • LLM fine-tuning & applied research — currently fine-tuning models and building agents for pathogen detection & diagnostic reporting
  • Full-stack product engineering — taking ideas from prototype to production on the web

🚀 Shipped Products

Agentic conference management — automating conference organization & operations end-to-end.

Collective-intelligence infra where independent AI agents interact, negotiate & converge.

AI Chief Growth Officer — automates growth marketing for D2C & e-commerce brands.

📌 Featured Projects

Project What it is
portable-triton-playbook Cross-platform Triton kernel tuning playbook — correctness-aware per-platform autotuner, W4A8 grouped GEMM MoE techniques (xcoal / widedot / vecw / predeq) · FlagOS 48h Kernel Bounty, T1 — 🏆 Best On-Site Award (34.60, 7/7 platforms)
uiniq · live Voice-first English speaking-practice companion — real-time voice agent (streaming ASR→LLM→TTS, barge-in), phoneme-level scoring, learner memory & SRS, WeChat native voice
hife-w8a8-quantization HiFloat8 W8A8 PTQ/QAT for Wan2.1-T2V-14B · IEEE ICME 2026 Low-bit LLM Quantization Challenge, W8A8 Training track · 🥇 1st as low team (76.13, $3,000)
QiNiuMagicRole AI character voice chat → one-click interview-podcast export · Next.js 14 + FastAPI + Grok-4 + GPT-SoVITS
CookNow 元启视界 AI Vibecoding Contest · 🥈 2nd place · also built DreamWeaver, an AI dream-journaling app
hare-ui Vue/TS component library (CLI · docs · tests) · ByteDance Youth Camp Frontend, 8th place · 超级码力奖

🏅 Awards

  • FlagOS 48h Kernel Bounty Challenge · Beijing (BAAI Conference 2026) · T1 w4a8_group_gemm_moe — 🏆 Best On-Site Award (34.60 avg speedup, 7/7 platforms) (portable-triton-playbook)
  • IEEE ICME 2026 Low-bit LLM Quantization Challenge · W8A8 Training track — 🥇 1st place as low team (76.13, $3,000)
  • Moore Threads MUSA Developer Challenge — 2nd place (Moore Threads AIBook)
  • 元启视界 AI Vibecoding Contest — 2nd place (CookNow)
  • 5th ByteDance Youth Camp · Frontend track — 8th place, 超级码力奖 (hare-ui)

🛠️ Stack

Python LangChain LLM APIs PyTorch TypeScript Next.js FastAPI Vue

📊 Stats

🐍 Contribution

contribution snake

Pinned Loading

  1. CookNow CookNow Public

    TypeScript

  2. dreamweaver dreamweaver Public

    TypeScript

  3. hife-w8a8-quantization hife-w8a8-quantization Public

    🥇 1st place (W8A8 Training track), IEEE ICME 2026 Low-bit LLM Quantization Challenge — Boundary-Protection W8A8 HiFloat8 quantization for Wan2.1-T2V-14B

    Python

  4. uiniq uiniq Public

    Python