Portrait of Jinsong Zhou

Jinsong Zhou周劲松

Ph.D. candidate in Artificial Intelligence at HKUST(GZ).

Building AI agents and multimodal generative systems.

Email GitHub Google Scholar

About me

AI agents, multimodal reasoning, and generative systems.

Jinsong Zhou (周劲松) is a Ph.D. candidate in Artificial Intelligence at the AI Thrust, Information Hub of The Hong Kong University of Science and Technology (Guangzhou), advised by Prof. Yingcong Chen. My work focuses on AI agents, multimodal reasoning, and generative systems. I am currently an AI Agent Solution Architect Intern at NVIDIA. Previously, I co-founded StoryVerse and worked as a Risk Control Algorithm Intern at ByteDance. I received my B.S. in Statistics from Southern University of Science and Technology.

Currently
AI Agent Solution Architect Intern, NVIDIA
Based in
Guangzhou, China
Advisor
Prof. Yingcong Chen

Experiences

From research labs to startups to industry.

  1. NVIDIA

    AI Agent Solution Architect Intern

    May 2026 – Present
    • Built cross-platform Agent Harnesses for expert workflows across DevRel, robotics, and AI coding agents.
    • Packaged repetitive NVIDIA DevRel workflows into 10 installable Agent Skills and delivered an internal Harness compatible with Codex, Claude Code, Cursor, and other major coding agents.
    • Converted expert-operated Sim2Real workflows—including data import, robot modeling, simulation tuning, validation, and configuration export—into a reproducible human-in-the-loop Agent Harness.
    • Built NIMHub, a reliable NVIDIA model integration layer supporting natural-language discovery, selection, multi-model composition, and code generation across 208 Build/NIM models, with direct invocation code for 163 models.

  2. StoryVerse

    Co-Founder

    Aug 2025 – Mar 2026
    • Co-founded an AI-native New Hollywood studio delivering end-to-end cinematic production for global audiences.
    • Helped close a $1M angel round at a $10M valuation.
    • Built and led a six-person cross-functional team across product, UI/UX, frontend, backend, and machine learning; productized the multi-agent cinematic pipeline and launched the Beta, supporting 1,000+ concurrent users.

  3. ByteDance, Douyin

    Risk Control Algorithm Intern

    May 2025 – Dec 2025
    • Built a LangGraph-based multi-agent risk assessment system from scratch, orchestrating 30+ MCP Tools to aggregate multi-dimensional user data for autonomous violation detection and explainable evidence generation.
    • Built an adversarial image synthesis engine and 3,000+ Function Tool call-chain samples; trained Qwen3-VL-8B with SFT and GRPO to jointly optimize tool selection and semantic reasoning, producing ImpText-Reader and the ICML 2026 paper ImpText.

Publications

Video generation, implicit text reasoning, and general-purpose generation.

All publications

5 papers · 53 citations · synced from Google Scholar. See all

Open Source

Plugins for the DeepSeek Harness.

safe-find-dsh-plugins

A DeepSeek Harness plugin that discovers task-relevant plugins and runs a security scan before installation.

View on GitHub

dsh-html-canvas

A DeepSeek Harness plugin that turns AI-generated HTML into a click-to-edit canvas beside the chat.

View on GitHub

Education

Honors & Awards

  • Full Postgraduate Scholarship HKUST(GZ)
    Sep 2024
  • First-Class Outstanding Student Scholarship (Top 5%) SUSTech
    Oct 2023
  • Study Abroad Scholarship (¥80,000) SUSTech
    Dec 2022

Skills

Coding Agents

  • Claude Code
  • Codex
  • Cursor
  • DeepSeek Harness

Programming

  • Python
  • LangChain
  • LangGraph
  • DeepAgents

Languages

  • English (Fluent, IELTS 7.0)
  • Mandarin (Native)

Visitors

Where readers of this page come from. Red dots are today's visitors; blue dots are everyone before them.

Powered by MapMyVisitors · updates once a day.