AI company 10 AI agents · approval

Connecting people and AI.

We design AI stacks and build AI agents on top of them. DualJ is an AI company founded by engineers who have built and consulted on AI agent stacks in real-world work.

Example workflow

Slack → Kanban → PR → CI → Approval → Deploy

  1. 01 Request Work comes in through Slack
  2. 02 Break down The lead agent creates Kanban tasks
  3. 03 Build 10 agents on code, docs and design
  4. 04 Verify CI and e2e tests
  5. 05 Approve People decide merges, deploys and posts
  6. 06 Deploy Ships to Cloud Run

#example workflow

lead → kanban.create("Update UI copy in ko·en·es")

engineer → github.openPR() · qa → check(390, 1366)

release → waiting_for_human_approval

10

AI agentsOur actual team setup

4

Model providersSwappable behind one adapter

3+

Multilingual serviceKorean · English · Spanish and more

What we do

We design, build and operate

Design the AI stack → build the agents → operate. We run this flow on our own products first, then design the same approach for other teams.

I Preparing for launch

AI services

We plan and build our own AI products. The first one built on the stack we designed is Yeppi.

Explore Yeppi
II

AI agent stacks

We design the stack by choosing models, tools and approval points, then build AI agents on it and operate them in real workflows.

Meet the agent team
III

AI adoption consulting

We help teams adopt AI in the same order: stack design, agent building, then operations.

See how it works

AI agent team

10 agents, people decide

They take work from Slack channels, split it into tasks on a Notion board and Kanban, and open GitHub PRs. People approve anything that is hard to undo.

Merge Deploy Payments Public posts → human approval required
  • Lead

    Project Lead

    Splits requests into Kanban tasks

  • Product planning

    Product

    Defines what to build and why

  • Engineering

    Engineer

    Writes code and opens GitHub PRs

  • QA

    Quality Assurance

    Checks tests and 390 · 1366 screens

  • Design

    UI Designer

    Works on screens and the design system

  • Growth marketing

    Growth

    Prepares outreach and growth

  • Business operations

    BizOps

    Keeps business operations on track

  • Customer support

    Support

    Organizes user inquiries

  • Release management

    Release

    Prepares deploys and waits for approval

  • Local model

    Qwen

    Takes on work with a locally run model

How work flows

Six steps from request to deploy

  1. 1

    Request Slack

    A task is posted in a Slack channel.

  2. 2

    Break down Kanban

    The lead agent creates tasks on Kanban.

  3. 3

    Build GitHub PR

    Agents produce code, docs and design.

  4. 4

    Verify CI

    CI runs, then e2e review.

  5. 5

    Approve Human

    A person reviews and decides on merge and deploy.

  6. 6

    Deploy Cloud Run

    Only approved changes ship to the service.

Kanban

Example workflow

In progress2

Split Shorts rendering job

Engineering

Tidy course page spacing

Design

Verify1

Apply Spanish copy

QA review

Awaiting approval

PR merge · Cloud Run deploy

Human's turn

Model adapter

Swap the model, keep the code

Feature code never calls a provider SDK directly; it goes through the adapter. So switching providers leaves feature code unchanged.

Feature code

course = ai.generate(
  task="travel_course",
  lang="es"
)

Model adapter

One interface

  • Gemini APIGemma 4
  • OCI Generative AIOracle Cloud
  • ClaudeAnthropic
  • Qwen 3.6Ollama · local

AI adoption consulting

We work this way first

An agent team with human approval is how DualJ already works every day. We propose the same flow to other teams: stack design → agent building → operations.

Talk about adoption
  1. I

    Assess

    We look at your work and data and list what the AI stack needs.

  2. II

    Design

    We design the AI stack: the model adapter, agent roles and the points where people approve.

  3. III

    Build

    We build AI agents on that stack, including tool integrations and cloud deployment.

  4. IV

    Operate

    We monitor and keep improving.

What we build

Yeppi

Preparing for launch

Yeppi is our first AI service and is preparing for launch. You can preview it at yeppi.info.

  • Travel courses

    Plan a trip as a course of places in order.

  • Photos to Shorts

    Turn trip photos into a short vertical video, rendered with Remotion · HyperFrames · FFmpeg.

  • Multiple languages

    Yeppi is available in Korean, English, Spanish and more.

  • Instagram integration Coming soon

    Instagram integration is planned and not available yet.

Preview Yeppi
Yeppi course page for a day in Busan, Gwangalli, with duration, number of stops, price and an Add to my trip button
Yeppi page showing sample 15-second Shorts made from travel photos of Seoul, Busan and Jeju
Yeppi home page displayed in Korean

Screenshots of yeppi.info; photo credits as shown on yeppi.info.

AI & Cloud Stack

What we actually use

Technologies we use in our own projects.

AI models

Model providers are swappable behind one adapter.

  • Gemini API (Gemma 4)
  • OCI Generative AI
  • Claude
  • Qwen 3.6 (Ollama)

AI agent stack

A team of 10 AI agents plans, builds, tests and reports, and people approve.

  • Hermes multi-agent
  • Slack
  • Notion
  • Kanban
  • MCP

Media AI

Photos become short vertical videos automatically.

  • Remotion
  • HyperFrames
  • FFmpeg

Google Cloud

Serverless apps, jobs and keyless deployment.

  • Cloud Run
  • Cloud Run Jobs
  • Artifact Registry
  • Secret Manager
  • Cloud Storage
  • Cloud Load Balancing
  • Workload Identity Federation
  • Firebase Hosting

Data & engineering

Databases, APIs, web apps, tests and CI.

  • Supabase (Postgres)
  • FastAPI
  • Next.js
  • Playwright
  • GitHub Actions

How we build

Three principles

i

People first

AI should make the people behind it more visible, not less.

ii

AI that actually runs

We ship real services, not demos, and we operate what we build.

iii

Small, fast and honest

We start small, move fast and say clearly what our AI can and cannot do.

“A future where AI connects people with experiences, places and each other.”

About

DualJ (듀얼제이) builds AI-powered services. We believe AI should widen the connections between people, not replace them. We have taken AI all the way into real services, from model integration and agent design to cloud operations, and we started DualJ small, and boldly, on that experience.

Team

Youngkyun Jung

Representative · Full-stack engineer

AI agent team

Working together under human approval

Contact

Questions about services, agent stacks or AI adoption consulting

contact@dualj.ai

We take inquiries by email.