AI Systems

Build the workflow.
Scale the craft.

I build and implement AI workflows that connect design intent with production tools. At Dongfeng, I introduced an MCP-orchestrated pipeline, taught it through live demonstrations and helped teams adopt it. Concept-video production accelerated 10–20×, with one designer running the workflow end-to-end.

Advanced workflows.
Capability the team can use.

My work combines AI orchestration, product research and design judgment. The Hunan University partnership used LLM-assisted synthesis alongside 1.78M+ user records and fieldwork across 11 cities. The examples below show character development, motion, brand exploration and a prompt-building tool, with concept studies distinguished from production applications.

Character systems.

Distinct projects. Clear identity and behavior.

Panda — Character turnaround and views

Dongfeng/Huawei.
A character with clear rules.

Panda identity, facial refinement, expressions and poses developed as a consistent reference for scenes and interfaces.

Explore the system
Lantern Lion — selected pose system

M-HERO Model B.
A distinct character language.

Lantern Lion pose development and projection studies connect character expression with the vehicle experience.

Explore the system
Drive-mode character system

M-HERO M18-3.
Character through graphics.

Drive-mode characters and a zodiac Easter egg turn functional states into a recognizable visual system.

Explore the system
Voice assistant — character identity

X700.
A voice assistant with identity.

One character adapts to expressions, app states, drive modes and seat profiles without losing its identity.

Explore the system

Motion.

A controlled starting frame. A finished sequence.

The starting frame · Scene and composition

Set the scene.
Direct the action.

I establish the composition, character positions and lighting before defining the camera behavior and movement for the animation.

Explore the system
Finished motion · 10 seconds · Play with sound

A clear path
from frame to film.

I carry the scene into motion, review character and background continuity, then finish the sequence for its intended screen.

Explore the system

Brand strategy.

An AI-assisted customer concept. A proposed visual direction.

The customer · Lifestyle and context

Explore the customer’s world.
Define the direction.

For Dark Current, I used AI to explore a proposed customer profile and a brand direction inspired by urban evening life. The profile is a concept to validate through research.

Explore the system
The concept · One geometry, two states

One geometry.
Two states.

The day and night interface concepts translate the brand direction into a shared geometry, with different expressions of light and color.

Explore the system

Tools & workflows.

A reusable visual brief. A tangible result.

The tool · Reference, subject and prompt

Build the tool.
Make direction reusable.

I built Object & Style Prompt Studio to create production prompts around a visual reference, an object and composition settings.

Explore the system
The output · Independent product exploration

One visual direction.
Consistent results.

The reference and prompt guide an independent helmet study, applying the selected material language and blue-purple lighting to the object.

Explore the system

Build the workflow.
Teach the capability.

The concept-video workflow moved from a multidisciplinary team working through sequential handoffs over 2–4 weeks to one designer producing the sequence in one day. Measured acceleration was 10–20× for that workflow; brief, review and finishing requirements affect timing.

I designed the teaching to build capability colleagues could carry into their own projects: understanding the workflow, choosing suitable methods and judging the result.

I connected AI models with After Effects, Premiere, Blender and Unreal, alongside generative image, video and sound tools. The workflow spans the brief, concepts, storytelling, frames, prompts, images, video and sound. Teaching through live tasks gave colleagues a method they could adopt and implement in their own work.

I structured the AI teaching around model fundamentals, reusable LLM templates and practical image generation, followed by storyboarding, video, looping animation and audio. Live demonstrations connected the curriculum with the production workflow.

I paired the speed with confidentiality rules, review of AI-generated claims and expert control of craft and correctness. Shared references, motion presets and sound systems kept the design language consistent as the team explored more directions.

Applied AI teaching at Dongfeng: connecting model fundamentals with practical creative workflows.