Saltware

AI Service

We offer enterprise chatbot services powered by AI to automate business processes.

Scroll

Fitsapie, the AI brand by Saltware

FitSapie is an AI brand independently developed entirely in-house by Saltware, consisting of AI-based chatbot (Sapie-Bot), product
recommendation (Sapie-Reco), and large language model (Sapie-LLM) services. In particular, the Korean language model Sapie, developed
in-house, ranked No.1 in the open-source category of the Tiger Leaderboard, which comprehensively evaluates Korean language
proficiency. This achievement proves that Saltware possesses advanced capabilities in data processing and model configuration.

Service 01
Sapie-Bot

LLM-based
AI chatbot
Service 02
Sapie-Reco

Hyper-personalized product
recommendations based on
user behavior patterns
Service 03
Sapie-LLM

Custom enterprise AI
model development

01.Sapie-Bot

Sapie-Bot is Saltware’s AI chatbot service based on LLMs.
It supports a 24/7 Q&A channel and a hyper-personalized work assistant,
addressing various customer needs and maximizing productivity and efficiency.

Sapie-Bot, what are its key strengths?

AI response system with enhanced accuracy and reliability
▪ Hallucination minimization ▪ Application of internal data via various data formats
Customer environment optimization
▪ Custom deployment in cloud, on-premise, or hybrid environments ▪ Fully customizable for customer-specific domains and tasks
Effortless scalability
▪ Providing core features via API for easy service expansion ▪ Seamless integration with legacy systems and databases
Service Architecture / Dashboard
▪ Store statistical data ▪ Store conversation history

Service Architecture

Advanced RAG

Minimizes hallucinations by providing accurate answers based on internal data.
Experience a fully customized RAG system that supports various data formats, optimized
retrieval strategies, and advanced chunking methods tailored to your needs.

02. Sapie-Reco

Sapie-Reco is a total recommendation service that offers a new shopping
experience by understanding user intent and behavioral context.
It generates rich and creative product themes based on product data,
enabling innovative recommendation beyond traditional methods.

Sapie-Reco, What are its key strengths?

Reduced development and operational resources
▪ Simple recommendation system development and operation using LLMs instead of complex algorithms ▪ Up to 700% improvement in operational efficiency compared to existing systems
Enhanced recommendation performance
▪ Up to 3% improvement in standard environments, more than 2x accuracy improvement in cold-start environments ▪ Improved purchase conversion rates and user engagement time when combined with existing recommendation systems
Improved customer shopping experience
▪ Hyper-personalized recommendations for a tailored shopping journey ▪ Rich and creative categorization that increases customer satisfaction

Service Architecture

03. Sapie-LLM

Sapie-LLM is Saltware’s proprietary LLM that enables domain-specific services
while completely preventing data leakage, offering a secure generative AI solution.
Backed by Saltware’s customization capabilities that ranked No.1 in the open-source category on the Tiger Leaderboard,
Sapie-LLM delivers a truly “Fit” model tailored to each customer.

Sapie-LLM, What are its key strengths?

Domain-specific customization
▪ Fine-tuning with customer data enables domain-specific services
High-level security
▪ Closed network environment for complete prevention of data leakage
Customizability and system integration
▪ Model customization service powered by leaderboard-proven technology ▪ API interface provided and modular structure enables high system integration

Performance Benchmark

Reference

1. Sapie-Bot Case: 24/7 Customer Q&A Chatbot for a Korean University

Y University is located in the Seoul metropolitan area and is known for its high employment rate.
It operated a KakaoTalk channel to communicate with students and parents, but since inquiries were handled manually,
responses were inevitably delayed during peak periods such as the admissions season, significantly reducing operational efficiency.
To improve both customer satisfaction and work efficiency, the university decided to adopt generative AI.
Saltware handled the entire process required for implementing and operating the generative AI service, including building a cloud infrastructure tailored to the client’s service environment, setting up an automated RAG pipeline, selecting a suitable model, and tuning prompts.
Even after the system handover, Saltware provided comprehensive technical support to ensure stable operation.
To prevent the delivery of incorrect information caused by hallucination, Saltware also provided advanced prompt engineering and customized admin tools for response management.
The chatbot was integrated with the university’s existing KakaoTalk channel and also deployed as a floating popup on the official website,
maximizing the use of existing systems and minimizing unnecessary development.
As a result, Y University reduced the manpower required for Q&A during the admissions season by more than 50%.
Students and parents were able to get answers to their questions at any time through the 24/7 chatbot, significantly improving service satisfaction.

2. Sapie-Reco Case: Hyper-Personalized Recommendation Service for Company B

Company B operates an online shopping mall that is especially popular among younger customers as a place to buy unique, handcrafted products.
In addition to its existing recommendation algorithm, the company decided to adopt generative AI in order to build a recommendation service that could suggest products in a fresher and more creative way for each season.
To suit the nature of the platform, where numerous new products are registered regularly, Saltware focused on building an automated pipeline to structure and utilize product and user information.
Saltware developed a pipeline that automatically performs clustering and creates a vector database when a product is registered, and another pipeline that uploads customer data and feeds it into prompts.
UI/UX was also customized for both the chatbot and the web page according to the client’s requirements.
The new recommendation service built into the customer platform showed meaningful results not only in internal testing but also with real users.
In particular, it effectively addressed the cold start issue of the existing recommendation system.
It also generated high-density recommendation themes in a much shorter time compared to human operation, improving customer satisfaction and delivering various operational efficiencies across the platform.

3. Sapie-LLM Case: Internal Policy Chatbot for Company S

Company S is an IT firm with approximately 130 employees.
Although internal policies, operational guidelines, and department contacts were published on the company’s internal platform, the operations support team was spending considerable time responding to employee questions about them.
To reduce this resource drain, Company S decided to implement a generative AI-based chatbot.
Because the company’s internal data included sensitive information, such as employee contact details and documents that must not be leaked externally, Company S wanted the service to operate in a closed network. They chose to build their service using Sapie-LLM.
Saltware fine-tuned Sapie-LLM using the company’s data so that the model could better understand terminology and context used in the IT domain.
At the same time, Saltware collected and organized data scattered across the company’s internal platform to build a RAG system, connected it to the model, and implemented the chatbot system within the company’s platform.
To support future use of multiple chatbots across different business areas, Saltware also provided Sapie Studio—enabling the client to easily generate and manage multiple chatbots and their datasets.
As the usage of the internal policy chatbot increased, the workload of the operations support team was significantly reduced, and the ability to immediately respond to policy-related inquiries improved overall work efficiency across the company.