Kakao i Dataverse

Overview

AI Data & Contents Management SaaS Platform

Company

Role

  • Product Manager &
    Lead Product Designer

Deliverables

  • Product Policy

  • UX Design

  • API Architecture

  • Data Report

Project Date

  • Mar 2022 - Aug 2023
    18 months

Kakao i Dataverse is a SaaS platform for managing content and data used in AI-powered services. To develop services using Large Language Models (LLMs), it is crucial to leverage Retrieval-Augmented Generation (RAG) technology. This involves integrating metadata-based search and vector search to retrieve relevant content and data, which then serve as the foundation for AI-generated outputs.

Kakao i Dataverse was designed as the next generation of bot services, evolving from B2B2C projects. It goes beyond the limitations of traditional rule-based bots by generating responses from data and content, enhancing service completeness.

The platform aims to enable non-technical administrators to operate and maintain AI services with minimal engineering effort.

Context & Background

The impact of Generative AI and the rising demand for AI transformation

With the rise of Generative AI, led by ChatGPT, many users now expect bots to generate answers dynamically, setting a new standard for satisfaction.
This shift has highlighted the limitations of traditional rule-based NLP bots that rely on intent and entity recognition. As a result, there is a growing need for an effective transition from conventional bots to generative AI-powered chatbots, along with easier maintenance solutions.

Hyphothesis : HMW Approach

How might we enable bot managers to easily manage, operate, and improve both legacy bots and new generative AI-based bots?

We explored how enterprise administrators using the existing Kakao i Connect (a B2B bot builder SaaS platform) could seamlessly transition to generative AI-powered bots.

To define the MVP features, we conducted stakeholder interviews with LLM researchers from our AI Lab. From a technical perspective, we decided to build the generative engine using KoGPT, a proprietary Korean language model based on the open-source GPT-2.

To ensure the completeness of our RAG (Retrieval-Augmented Generation) service, we collaborated with Daum’s search technology team. This involved integrating a metadata-based index search engine and incorporating Vector Search, which leverages our proprietary search database. These integrations formed the technical foundation of our solution.

Following this, we planned user research involving interviews with bot managers and end users to refine the MVP’s user experience and functionality.

User Research

  • Kakao AI Lab LLM Model Researcher

  • Kakao Search Engine Lead Engineer

  • Kakao AICC Service Product Owner

  • NH Investment Chatbot Manager

  • Hey Kakao Customer Service Agent

5 Stakeholders Interviews

  • Integrated generative AI responses into the stock bot
    (Hey Kakao and Kakao Work)

  • Implemented generative AI responses functionality in AICC SaaS.

  • Implemented AI-driven answers for NH Investment & Securities' stock bot.

3 Proof-Of-Concept Projects

  • Codeless data and content storage with real-time integration for bots and services.

  • Provision of a private repository, allowing bots to execute processes while exempting data from model training.

  • Real-time analytics to track user data flow, identify drop-off points, and monitor conversions.

  • Role-based access control (CRUD permissions) for different team members.

5 Defined User Needs

Outcomes

Kakao i Dataverse

We launched Kakao i Dataverse

A Low-Code based SaaS Platfrom for storing, managing and integrating Data&Content, with the potential to expand into Generative AI

Service Flow

Kakao i Dataverse can be directly served as a REST API for user’s own service

Kakao i Dataverse’s key features include data & content storage management, search engine configuration, and generative AI management. By leveraging these capabilities, it supports the implementation of generative AI bots using Kakao’s on-premise LLM model, as well as OpenAI and Meta LLaMA models.

Integrated Dashboard

Integrated Dashboard enables managers to easily monitor key status

Effortlessly track all data and content in one view

Monitor real-time data status with card views for each Category. Quickly access detailed data with a single click.

Easily check key status changes at a glance

Check data and content sync expiration through lists and status indicators. Monitor frequently accessed search keywords and data in real time for quick operational response.

UX Flow-based Data Analytics

Analyze user data flow to deliver a seamless service

Visualize user flow at a glance with a Sankey chart

UX Flow-based analytics feature visualizes the depth of user-accessed data through a Sankey chart, providing a clear view of content usage patterns and drop-offs.

Visual charts and graphs tailored to each Insight

Charts and graphs help provide clear insight analysis

To intuitively understand trends in large datasets, charts and graphs tailored to each insight are essential. Supports line graphs, pie charts, bar charts, and more for easy analysis.

Search Index & Meta Data Management

Kakao i Dataverse enables the independent creation, maintenance, and customization of each service’s search engine with simple UI/UX

Create custom search indexes & rules to operate your own search service

Keyword and morphological data can be easily uploaded in bulk with a simple click of a button. Search rules can be effortlessly created by selecting representative datasets and adding bubbles.

Generative AI Management

Retrieval-augmented generation(RAG) AI services can be easily created, integrated, and operated through a simple process

Content upload for AI, prompt settings, and search engine integration are possible

With just a few clicks, service manager can add generative AI features, convert existing bots, and integrate content with AI models. Apply search indexes and vector search rules easily, and enhance responses with prompt instructions. Data annotation ensures sustainable operation.