LAVENDER PROJECT · COMPANY

Lavender Project

AI that understands the work. Software built for operations.

An independent software business in South Korea. We develop web services and system integrations that reduce repetitive work while keeping people in control of important decisions.

01 / SERVICES

From a useful feature to a working service.

We extend existing systems and build new web services. Every engagement starts with a defined scope and clear acceptance criteria.

Web development

Customer-facing websites, admin tools and backend APIs, including feature development, maintenance and troubleshooting for existing services.

Systems & API integration

Connecting orders, payments and notifications to reduce repeated data entry. Integrations and exception handling are scoped around existing operations.

AI agents & chatbots

Designing and evaluating AI features grounded in business rules and company knowledge. The two use cases below are currently being prototyped.

Development services & pricing (Korean)

02 / IN DEVELOPMENT

Two workflows we are building.

These diagrams illustrate our development direction. Both prototypes are in development; they do not represent launched products or customer deployments.

Prototype in development

An e-commerce agent with human approval

We are building an agent that reviews order data and refund policies, identifies orders that may need a refund, and explains its recommendations to an operator. Backend updates are designed to happen only after the operator reviews and explicitly approves the proposed action.

Recommendation and execution are separate steps. An unapproved order is not processed.

E-commerce agent concept: review orders and refund policies, recommend candidates with reasons, request operator approval, then update approved records. No processing without approval.
Concept 01 — Backend changes are designed to follow explicit human approval.

Prototype in development

Customer support grounded in company knowledge

We are developing a chatbot that retrieves relevant information from each company’s product documentation, policies and FAQs. Retrieval-augmented generation (RAG) provides that source material to a language model to help explain answers in the company’s context.

The intended workflow includes source references and a handoff to a person when the available information does not support an answer.

RAG support chatbot concept: search company documents and FAQs for relevant information, answer using retrieved sources, show references and hand off unsupported questions to a person.
Concept 02 — Retrieve company information, explain with sources, and hand off when needed.

03 / FOUNDER

Built by Kichang Jeong.

Kichang Jeong runs Lavender Project, working across frontend development, backend systems and service operations. This website also hosts his personal engineering blog, documenting development decisions, experiments and lessons learned.

What would you like to improve?

Tell us about your current system, the work you want to simplify and your timeline. We can start by defining a practical scope together.

ceo@kichang.info

An inquiry does not create a contract or a payment obligation.